Chak-Hau Michael Tso B.S. M.S. Prof. Andrew Binley · Roses. I thank my family and friends around...
Transcript of Chak-Hau Michael Tso B.S. M.S. Prof. Andrew Binley · Roses. I thank my family and friends around...
Enhancing the information content
of geophysical data for nuclear site
characterisation
by
Chak-Hau Michael Tso B.S. M.S.
Supervisor:
Prof. Andrew Binley
Thesis submitted in partial fulfilment for the
degree of Doctor of Philosophy in Environmental Science
September 2019
Lancaster Environment Centre
Abstract
i
Enhancing the information of geophysical data for nuclear site characterisation
Chak-Hau Michael Tso
A thesis submitted for the degree of Doctor of Philosophy, Lancaster University
September 2019
Abstract
Our knowledge and understanding to the heterogeneous structure and
processes occurring in the Earth’s subsurface is limited and uncertain. The above is
true even for the upper 100m of the subsurface, yet many processes occur within it (e.g.
migration of solutes, landslides, crop water uptake, etc.) are important to human
activities. Geophysical methods such as electrical resistivity tomography (ERT) greatly
improve our ability to observe the subsurface due to their higher sampling frequency
(especially with autonomous time-lapse systems), larger spatial coverage and less
invasive operation, in addition to being more cost-effective than traditional point-
based sampling. However, the process of using geophysical data for inference is prone
to uncertainty. There is a need to better understand the uncertainties embedded in
geophysical data and how they translate themselves when they are subsequently used,
for example, for hydrological or site management interpretations and decisions. This
understanding is critical to maximize the extraction of information in geophysical data.
To this end, in this thesis, I examine various aspects of uncertainty in ERT and develop
new methods to better use geophysical data quantitatively. The core of the thesis is
based on two literature reviews and three papers.
In the first review, I provide a comprehensive overview of the use of
geophysical data for nuclear site characterization, especially in the context of site clean-
up and leak detection. In the second review, I survey the various sources of
uncertainties in ERT studies and the existing work to better quantify or reduce them. I
propose that the various steps in the general workflow of an ERT study can be viewed
as a pipeline for information and uncertainty propagation and suggested some areas
have been understudied. One of these areas is measurement errors. In paper 1, I
compare various methods to estimate and model ERT measurement errors using two
Abstract
ii
long-term ERT monitoring datasets. I also develop a new error model that considers
the fact that each electrode is used to make multiple measurements.
In paper 2, I discuss the development and implementation of a new method for
geoelectrical leak detection. While existing methods rely on obtaining resistivity
images through inversion of ERT data first, the approach described here estimates leak
parameters directly from raw ERT data. This is achieved by constructing hydrological
models from prior site information and couple it with an ERT forward model, and then
update the leak (and other hydrological) parameters through data assimilation. The
approach shows promising results and is applied to data from a controlled injection
experiment in Yorkshire, UK. The approach complements ERT imaging and provides
a new way to utilize ERT data to inform site characterisation.
In addition to leak detection, ERT is also commonly used for monitoring soil
moisture in the vadose zone, and increasingly so in a quantitative manner. Though
both the petrophysical relationships (i.e., choices of appropriate model and
parameterization) and the derived moisture content are known to be subject to
uncertainty, they are commonly treated as exact and error‐free. In paper 3, I examine
the impact of uncertain petrophysical relationships on the moisture content estimates
derived from electrical geophysics. Data from a collection of core samples show that
the variability in such relationships can be large, and they in turn can lead to high
uncertainty in moisture content estimates, and they appear to be the dominating
source of uncertainty in many cases. In the closing chapters, I discuss and synthesize
the findings in the thesis within the larger context of enhancing the information content
of geophysical data, and provide an outlook on further research in this topic.
Executive summary for the nuclear industry
iii
Executive summary for the nuclear industry
Uncertainty in the subsurface characterisation of nuclear sites poses significant risks in
terms of operational cost and environmental protection. Improved knowledge of the
uncertainty of subsurface properties and processes is needed in order to enhance risk
mitigation. Geophysical methods, such as electrical resistivity tomography (ERT),
provide a cost-effective way to delineate variations in subsurface properties and
monitor subsurface processes, however, the uncertainty in the results from such
methods is often overlooked. A recent successful time-lapse ERT field trial conducted
at Sellafield's Magnox Swarf Storage Silo (MSSS) highlights the potential of these
methods [1] by showing 3D resistivity variations over time due to saline tracer injection.
This PhD project explores various ways to better exploit information from ERT and to
track the associated uncertainty in subsurface characterisation. This includes better
understanding of the ERT data, and incorporating ancillary data sources to the ERT
analysis.
We have studied the error structure in ERT data and proposed a new error model for
geophysical measurements, which shows improved ERT inversion results and
uncertainty estimation [2]. Recently, we have shown that there exists large variability
in field petrophysical relationships and have developed a workflow quantifying pore
water states (e.g. soil water content) derived from ERT. Even though different
petrophysical relationships give consistent estimates of the change in total moisture,
the estimates have large uncertainty bounds [3]. Our study also illustrates the joint use
of coupled hydrogeophysical modelling and data assimilation to effectively estimate
flow and transport properties in leak plumes. Our method proposes a range of
hydrological models and then constrains them with time-lapse ERT data through data
assimilation. The advantages of this method includes the flexibility to incorporate prior
hydrogeological information and the ability to estimate flow and leak parameters of
interest directly. The ensemble of hydrological model estimates also readily provides
useful metrics for site management decisions, e.g. mass flux and mass discharge at any
location or area within the model domain.
Executive summary for the nuclear industry
iv
We have applied the above methods to the data collected from the Sellafield field trial
and other sites. Overall, our work addresses the needs of the Nuclear
Decommissioning Authority (NDA) by offering a suite of methods that can make
geophysical methods more reliable and informative for site characterisation.
Systematic application of ERT at NDA sites should contribute to a reduction in costs
and risks in managing NDA's contaminated land portfolio.
References:
[1] Kuras et al. (2016) Science of the Total Environment.
DOI: 10.1016/j.scitotenv.2016.04.212
[2] Tso et al. (2017) Journal of Applied Geophysics. DOI: 10.1016/j.jappgeo.2017.09.009
[3] Tso et al. (2019) Water Resources Research. DOI: 10.1029/2019WR024964
Table of Contents
v
Table of Contents
Contents
Abstract ...................................................................................................................................... i
Executive summary for the nuclear industry ..................................................................... iii
Table of Contents .................................................................................................................... v
Acknowledgment ................................................................................................................... vi
Declaration ............................................................................................................................ viii
List of Figures ......................................................................................................................... ix
List of Tables ......................................................................................................................... xvi
List of Acronyms ................................................................................................................. xvii
1. Introduction ................................................................................................................... 18
1.1 Background ............................................................................................................ 18
1.2 Objectives and aims .............................................................................................. 18
1.3 Outline .................................................................................................................... 20
2. Geophysical methods for nuclear site characterisation ........................................... 23
3. Sources of uncertainties in electrical resistivity tomography (ERT): a review ..... 59
4. Paper 1: Improved characterisation and modelling of measurement errors in
electrical resistivity tomography (ERT) surveys ............................................................. 115
5. Paper 2: Integrated hydrogeophysical modelling and data assimilation for
geoelectrical leak detection ................................................................................................ 166
6. Paper 3: On the field estimation of moisture content using electrical geophysics—
the impact of petrophysical model uncertainty .............................................................. 210
7. Discussion summary .................................................................................................. 248
8. Conclusions and recommendations ......................................................................... 255
8.1 Conclusions .......................................................................................................... 255
8.2 Future work ......................................................................................................... 256
Bibliography ........................................................................................................................ 257
Appendix 1: Instructions on using PFLOTRAN-E4D .................................................... 300
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method .................................................................................................................................. 303
Appendix 3: Annotated bibliography for related textbooks ......................................... 312
Vita ........................................................................................................................................ 314
Acknowledgment
vi
Acknowledgment
This thesis would not be possible without the help of many people and
institutions.
My PhD experience has benefited from much input from my doctoral
supervisor Andrew Binley. I thank him for challenging me with new ideas and
opportunities and always providing insightful reviews of my work. I am grateful for
his reassurance when I felt my work was not going anywhere. He has shown great
patience, kindness, and forgiveness throughout my PhD. I also thank the members of
the Binley group, Paul Mclachlan, Guillaume Blanchy, Tuvia Turkeltaub, Jimmy Boyd,
Qinbo Cheng, John Ball and visitors to the group for their friendship and creating a
stimulating and collaborative research environment. I also thank Lai Bun Lok
(Engineering) and Andrew Curtis (Edinburgh Universtiy) for thoroughly examining
this thesis and providing helpful comments during my viva.
My co-supervisor Oliver Kuras recruited me for this incredible opportunity. I
thank him for his warm welcome, constant support and helpful comments throughout
my PhD. I thank members of the BGS GTom team, especially Jon Chambers, Paul
Wilkinson, and Seb Uhlemann for their helpful suggestions to many parts of my work.
This PhD is supported by a Faculty of Science and Technology studentship and
a UK Nucelar Decommissioning Authority bursary. I thank my industry supervisor
James Graham for his valuable experience in nuclear site characterisation and other
NNL staff for planning activities that has deepened my understanding of the nuclear
industry.
I am deeply grateful for the support and feedback from the wider
hydrogeophysics community, especially our meetings in AGU and webinars. I thank
my external collaborators Tim Johnson, Xingyaun Chen, Xuehang Song, Marco Iglesias,
Andrew Curtis, Erica Galetti, Yuanyuan Zha, Chin Man Mok and Barbara Carrera for
extra research opportunities that has broadened my PhD experience, and for hosting
me during my visits.
Acknowledgment
vii
I thank fellow members of the CEH Environmental Data Science Team, for their
understanding while I have been finishing up my PhD. I thank Lancaster University
Men’s volleyball team for three memorable seasons and opportunities to play at the
Roses. I thank my family and friends around the world for their love and
encouragement. I thank my friends Welson, Roy, Richard, and Xichen for sharing their
experience and encouraging me. I thank my parents Paul and Helen for always letting
me choose a subject that I am interested in to study, and my brother Matthew for his
companionship. I am indebted to my parents-in-law John and Melody, especially for
allowing me to take their daughter to England. I thank my wife and best friend,
Elizabeth, for her love, laughter, encouragement, patience, and sacrifices every day
and going through the emotions of the PhD with me. I also thank our baby daughter,
Jemimah, for big smiles to welcome “baba” home from work. Finally, I thank the
greatest Giver of all my Lord Jesus Christ. I thank Moorlands Church for faithful Bible
teaching and discipleship after we arrived in Lancaster. Thank you for growing my
understanding of the Bible and my desire to bring people to Christ.
“Seek the Lord while he may be found; call to him while he is near. Let the wicked one
abandon his way and the sinful one his thoughts; let him return to the Lord, so he may have
compassion on him, and to our God, for he will freely forgive.” (Isiah 55:6-7, Christian
Standard Bible)
Declaration
viii
Declaration
Except where reference or is made to other sources, I declare that the work in
this thesis is my own and has not been previously submitted, in part or in full, to any
institution for any other degree or qualification.
Chapter 4 (Improved characterisation and modelling of measurement errors in
electrical resistivity tomography (ERT) surveys, Tso et al. 2017) has been published in
the peer reviewed publication Journal of Applied Geophysics, as described in the
reference list below.
Chapter 6 (On the field estimation of moisture content using electrical
geophysics – the impact of petrophysical model uncertainty, Tso et al. 2019) has been
published in the peer reviewed publication Water Resources Research, as described in
the reference list below.
For both of the abovementioned chapters, the co-authors contributed to the
interpretation of results, design, or data provision of the studies and edited drafts. A.B.
conceived the study and I wrote the manuscripts, developed scripts, and performed
analysis. Chapter 2, 3, and 5 are prepared to be submitted to academic journals for
peer-review.
Tso, C.-H.M., Kuras, O., Wilkinson, P., Uhlemann, S., Chambers, J., Meldrum, P.,
Graham, J., Sherlock, E., Binley, A. (2017): Improved modelling and characterisation of
measurement errors in electrical resistivity tomography (ERT) surveys. Journal of
Applied Geophysics, 145, 103--119, DOI:10.1016/j.jappgeo.2017.09.009.
Tso, C.-H.M., Kuras, O., Binley, A. (2019): On the field estimation of moisture content
using electrical geophysics‐the impact of petrophysical model uncertainty. Water
Resources Research, 55, 2019, DOI:10.1029/2019WR024964.
Chak-Hau Michael Tso B.S. (Texas) M.S. (Arizona)
Lancaster University, UK
List of Figures
ix
List of Figures
Chapter 1
Figure 1 The ERT workflow showing the various stages of conducting ERT survey and
analysing its data. It also serves as a pipeline where information and uncertainty is
propagated along. The annotation shows the relation between the chapters in this
thesis and the workflow. ................................................................................................. 20
Figure 2 Snapshot showcasing coupled hydrogeophysical modelling using
PFLOTRAN-E4D. The 2D flow and transport model simulates tracer injection at an
injector in the upper left of the domain. Groundwater movement is towards an
extraction well to the lower right. An ERT imaging cell with 4 boreholes (20
electrodes each) is located at the centre of the domain. The animation shows that as
the conductive tracer migrate through the ERT imaging cell, there is a
corresponding increase in electrical conductivity in the ERT imaging cell. ............ 22
Chapter 2
Figure 1 Aerial view of the Hanford Site (Johnson et al., 2015a) .................................... 29
Figure 2 Estimated conductivity at the Hanford BY-Cribs (Johnson et al., 2010; Johnson
and Wellman, 2013) ......................................................................................................... 30
Figure 3 A screenshot for the browser-based integrated data platform SOCRATES for
the Hanford site. It serves as a centralized portal for all data collected at Hanford
Site. It includes a wide variety of tools, such as those for filtering data and exporting
data and model domain for flow and transport modelling. ...................................... 35
Figure 4 An illustration of the iterative decision analysis framework (Paté-Cornell et
al., 2010). ............................................................................................................................ 46
Chapter 3
Figure 1 Various sources of errors and uncertainties propagate through the ERT
workflow (Binley et al., 2015; Tran et al., 2016; Truex et al., 2013). The workflow
begins with experimental design, where the objectives and details of the field
campaign are laid out. It progresses to data collection in the field using a data
acquisition system. Then the data is inverted to obtain results in a usable format for
interpretations and discussions. Finally, the findings are used for decision making
or prediction of future events. ........................................................................................ 62
Figure 2: Example of 3D borehole effects in 2D inversion reported in Nimmer et al.
(2008), which shows the 2.5D inversion results from Slater et al. (1997). The heavily
fractured zone at around 25-m depth can be seen as a low resistivity contrast to the
background, but the large high resistivity (>2x105 Ω m) in the centre of the image
appears to be a result of the 2D resistivity model compensating for the low
resistivity along the borehole. ........................................................................................ 66
List of Figures
x
Figure 3 An outline of the Markov chain Monte Carlo (McMC) inversion algorithm.75
Figure 4 Uncertainty quantification in Bayesian inversion (Iglesias and Stuart, 2014).
The black dashed lines and red solid lines denote prior and posterior probabilities.
Essentially, the data drives an updating of input probabilities (e.g. model
parameters) and lead to an updating of model outputs, including quantities of
interest that are not directly observable from data. .................................................... 77
Figure 5 This synthetic example shows that by disconnecting the smoothness
constraint in a regularized ERT inversion, the fracture network (red and yellow) is
much better recovered (Robinson et al., 2015). Compared with the smoothness
constraint inversion, the smoothness disconnect case shows pronounced elongated
fractures and recover the very high conductivities along them. ............................... 84
Figure 6 A new Bayesian inversion method that jointly estimates both the interface of
the two units and the sub-unit resistivity variations (de Pasquale et al., 2019). The
approach estimates the resistivty field of both unit (assuming they span the entire
model domain) and their interface. The resultant field is obtained by combining the
two fields along the interface. ......................................................................................... 85
Figure 7 Illustration of different approaches for integrating multiple geophysical data
sets for a hydrogeological interpretation (Doetsch, 2011). Individual processing
invert each dataset invidually before interpreting them together qualitatively. Joint
inversion invert all available datasets together—explicit assumptions between
different process models are required. A constrained inversion use parts of the
inversion results from one inversion to constrain another......................................... 87
Figure 8 The ERT workflow as a pipeline for information and uncertainty propagation
is helpful for the experimental design of ERT surveys. It can be optimized to
minimize uncertainty and maximize the extraction of information. ........................ 90
Chapter 4
Figure 1 Synthetic problem for demonstration (a) Synthetic domain with a more
conductive layer near the surface and a resistive area between x = 15m and x = 20m.
The synthetic data from running a forward model in (a) is perturbed with 5%
Gaussian noise and then inverted by assuming (b) 10% linear error model (c) 5%
linear error model (d) 2% linear error model. Note that rms error is defined as 𝒊 =
𝟏𝒏(𝒐𝒃𝒔 − 𝒔𝒊𝒎)𝟐/𝒏, where obs and sim are vectors of observed/true and simulated
transferred resistances of length n respectively. Note that the convergence target for
all the inversions is a chi-squared statistic of 1. ......................................................... 122
Figure 2 (a) Comparison of stacking errors, repeatability errors, and reciprocal errors
for the Boxford dataset by plotting probability density functions. The PDFs of
reciprocal errors and repeatability errors are comparable to each other. The stacking
errors PDF, however, show very low mean and low variance. Using stacking errors
for measurement errors characterisation may lead to significant underestimation of
uncertainty and over-fitting of data. (b) Comparison of stacking errors, repeatability
List of Figures
xi
errors, and reciprocal errors for the Sellafield dataset. The PDFs for Sellafield show
greater variances than those for Boxford. Since a two-week repeatability cycle is
used, the repeatability errors are much greater than reciprocal errors. In general, the
stacking errors are more than an order-of-magnitude smaller than the reciprocal
errors, indicating there may be significant underestimation of errors if they are used
as error weights. The mean and standard deviation of each fitted normal
distribution is shown next to the legend. ................................................................... 134
Figure 3 Autocorrelation of (a) departure from the mean (as a measure of repeatability
errors) and (b) reciprocal errors for the 96 datasets collected continuously
continuously within 24h at the Boxford site. The number of lags is on the horizontal
axis (here 1 lag = 15 minutes). Each grey translucent line plots the autocorrelation of
one of the 516 ERT measurements as a function of lag. The red line denotes the mean
autocorrelation. For each autocorrelation plot, 96 datasets are considered. The
hashed region has insignificant correlation according to the critical Pearson’s test
(around ±0.2). .................................................................................................................. 136
Figure 4 Autocorrelation of (a) departure from the mean (as a measure of repeatability
errors) and (b) reciprocal errors for the 96 datasets collected continuously at the
Sellafield site encompassing the three injection periods (22/1/2013 – 3/11/2013). The
number of lags is on the horizontal axis (here 1 lag = ~2 to 3 days). Each grey
translucent line plots the autocorrelation of one of the 12481 ERT measurements as
a function of lag. The red line denotes the mean autocorrelation. For each
autocorrelation plot, 96 datasets are considered. The hashed region has insignificant
correlation according to the critical Pearson’s test (around ±0.2). Similarly, (c) and
(d) show the same for the long-term background monitoring period (i.e. no injection,
5/11/2013 – 31/3/2014). ................................................................................................... 136
Figure 5 Mean correlation coefficient of departure from the mean (as a measure of
measurement errors) and reciprocal errors for measurement pairs from the Boxford
dataset as a function of dipole separation multiplier n. For both departure from the
mean and reciprocal errors, mean correlation coefficients are distinctively higher for
measurements that share electrode(s) in their quadrupoles than the mean
correlation coefficients for all measurements, indicating by considering the effect of
using each electrode to make multiple measurements may improve error models.
Also, note that the reciprocal errors have strikingly lower correlation coefficients
than the departure from the mean. Note that electrode sharing only occurs in ~10%
of all pairs. ....................................................................................................................... 138
Figure 6 Comparison of fitting reciprocal errors of time-lapse data as (a) individual
datasets,fitting each dataset individually with a different LME model and (b)
longitudinal data, fitting all data with one LME model. The above shows that it is
much better not to treat errors as longitudinal data. ................................................ 140
Figure 7 Synthetic surface ERT experiments to demonstrate the performance of the
error models. For data involving 3 bad electrodes(marked by “X”), data is corrupted
List of Figures
xii
by 10% white noise while for the rest of the data 2% white noise is added. (a)
Inverted resistivity distribution using the 2% linear error model (b) Inverted
resistivity distribution using a 4.52% (obtained from the Koestel et al. (2008) method)
linear error model (c) Inverted resistivity distribution using the LME error model.
Note that the convergence target for all the inversions is a chi-squared statistic of 1.
........................................................................................................................................... 143
Figure 8 (a – c) Diagonal of resolution matrix for inversion using the following error
models for inverting the synthetic data corrupted by “bad electrodes”: (a) 2% linear
model (b) 4.52% linear model (c) LME model. (d - f) variance of element-wise log-
resistivity estimates using each of the error models obtained from Monte Carlo
experiments. The colour scale is the same for all three error models. Darker cells
indicate more similar model estimates between Monte Carlo estimates. (g - i) mean
model estimates from Monte Carlo experiments. The transparency is controlled
linearly the variance shown in (d – f). With model averaging, the mean estimates of
the three error models agree. It is noted, however, the deterministic results from
the LME model agrees the best with its model-averaged results. ........................... 144
Figure 9 Empirical model covariance matrix using the Monte Carlo uncertainty
propagation procedure and the following error models: (a) 2% linear model (b)
4.52% linear model (c) LME model. The size of the matrix is m x m, where m is the
number of model parameters. By comparing (a) and (b), it is shown that assuming
higher error levels, there is higher covariance between model parameters. With the
LME error model, the model covariance is the lowest. While the spread of high
covariance entries are quite even throughout the matrix, we can see that the spread
for (c) is quite uneven: generally, elements on the left of the domain have higher
spread. .............................................................................................................................. 145
Figure 10 Inversion results from Boxford using (a) linear error model for stacking
errors, (b) LME error model for stacking errors, (c) linear error model for reciprocal
errors. ............................................................................................................................... 146
Figure 11 (a) 3-D static deterministic inversion results from Sellafield on 5th February,
2013. Error weights are prescribed by fitting an LME error model. Black lines are
boreholes installed with electrodes. (b) The corresponding uncertainty estimates
obtained from Monte Carlo simulations, given by model standard deviation from
Monte Carlo experiments. (c) The corresponding coefficient of variation of Monte
Carlo model estimates. .................................................................................................. 147
Chapter 5
Figure 1 (a) Flowchart of the overall data assimilation framework used in this work.
More details are found in the subsections. (b) The goal of this framework is that
upon conditioning of geophysical data, the envelope of possible mass discharge
time series will become less uncertain. ....................................................................... 173
List of Figures
xiii
Figure 2 . (a) PFLOTRAN model domain for the Sellafield MSSS. The grey area is the
MSSS building, which is modelled as impermeable. The hashed area is the ERT
imaging cell consisting of four ERT boreholes. (b) A snapshot of the simulated tracer
concentration due to injection. (c) The corresponding distribution of electrical
conductivity within the ERT imaging cell obtained via petrophysical transform.
.......................................................................................................................................... 180
Figure 3 Estimation of leak location. (a) The true leak location is within the ERT array
(33.4534, -14.4303). (b) The true leak location is outside the ERT array (20, -10). In
both cases, the data assimilation framework successfully identified the true leak
location within a few iterations. ................................................................................... 183
Figure 4 Joint estimation of leak parameters: (𝒙, 𝒚) location, leak rate, and onset time.
(a) Violin plots showing the prior and posterior parameter distributions. The true
values are marked with an orange lines. The posterior parameter values collapse
around the true values (b) Prior and posterior tracer mass flux across the pre-
defined plane. All the posterior curves collapse to nearly the true curve (green).
Note that the sign of mass discharge denotes its direction across the plane......... 184
Figure 5 Joint estimation of leak and petrophysical parameters: the prior and posterior
parameter distributions are shown as violin plots. The true values are marked with
orange lines. .................................................................................................................... 185
Figure 6 (a) The estimation leak parameters under uncertain K values and logK
variance = 1.0. The violin plots show the prior and posterior parameter distributions.
The true value is marked with an orange line. (b) The estimation of leak parameters
at variance of log10(K) equal to 2, 3, 5, 7, 10 , while assuming the mean K values are
known exactly and the K field is isotropic and is of unit correlation length. The
violin plots show the posterior parameter distribution, while the true value is
marked with an orange line. ......................................................................................... 186
Figure 7 (a) Joint estimation of leak parameters and effective hydraulic conductivity.
The violin plots show the prior and posterior parameter distributions. The true
value is marked with an orange line. (b) Prior and posterior tracer mass flux across
the pre-defined plane. The true curve is marked in green in the posterior plot. .. 187
Figure 8 Setup of the tracer injection test at Hatfield (H-I2 is the injection borehole and
H-E1 to H-E4 are ERT boreholes) and the time-lapse resistivity images (iso-surfaces
are plotted for 7.5% reduction of resistivity relative to baseline) obtained from a
difference inversion of the ERT data (reproduced from Winship et al., 2006) ...... 189
Figure 9 (a) Parameter scatterplots showing pairs of parameter values for the Hatfield
example estimating leak and Archie parameters. The parameter symbols and units
are defined in section 3. Grey squares indicate prior parameter values while black
circles in date posterior values. The true leak parameters used in the field injection
experiment is indicated by red triangles. (b) The prior and posterior mass discharge
time series. The sign of mass discharge indicates the direction across the defined
plane. ................................................................................................................................ 192
List of Figures
xiv
Figure 10 Parameter scatterplots showing pairs of parameter values for the Hatfield
example estimating leak and Archie parameters and hydraulic conductivities. The
parameter symbols and units are defined in Table 3 and section 3. Grey squares
indicate prior parameter values while black circles indate posterior values. The true
leak parameters used in the field injection experiment is indicated by red triangles.
........................................................................................................................................... 193
Figure 11 Transfer resistance scatter plot between the observed and simulated data at
Hatfield. The simulated data uses parameter values listed in Table 3. .................. 195
Chapter 6
Figure 1 Moisture content 𝜽 estimation and petrophysical uncertainty propagation
workflow used in this study. Rectangles indicate model inputs or data, while ovals
represent modeling or analysis steps. We obtained synthetic ERT and 𝜽 data using
PFLOTRAN‐E4D. Then we inverted the ERT data and used the Eggborough cores
as different petrophysical models. They were passed through the moisture content
estimation and uncertainty estimation framework to obtain ERT ‐ estimated 𝜽 ,
which were compared against the 𝜽 data. ERT = electrical resistivity tomography.
........................................................................................................................................... 218
Figure 2 (a) Cumulative density functions of grain size distribution of Eggborough
cores and blocks. The legend shows the core or block ID. (b) Depth profiles of sand
(blue), silt (red), and clay (yellow) percentages for Eggborough cores. ................. 219
Figure 3 Archie's parameter estimation of individual Eggborough cores and blocks.
The predictions using the best estimate of the parameters are shown in solid lines,
while the 68% (i.e., ±1 standard deviation) confidence intervals are shown in dashed
lines. Note that the measurements are made at 𝝈𝒇 = 𝟏𝟎𝟎𝟎 𝛍𝐒 𝐜𝐦 − 𝟏. Note that 𝝆,
which is the dependent variable, is shown on the x‐axis. ........................................ 225
Figure 4 Summary of Archie model fits for the Eggborough/Hatfield cores and blocks.
Note that values correspond to 𝝈𝒇 = 𝟏𝟎𝟎𝟎 𝛍𝐒 𝐜𝐦 − 𝟏. The point label “synthetic”
is the “true” solution considered in the synthetic study in section 3.2. .................. 225
Figure 5 (a) Mean (log10) and (b) standard deviation (linear) of electrical resistivity for
Day 18 obtained from Monte Carlo runs of electrical resistivity tomography
inversion. (c) Extracted volume where there was a 5.5% reduction of resistivity
relative to baseline on Day 18. The purple cubes are electrode locations. ............. 227
Figure 6 (a) Total water volume within the extracted volume (with uncertainty bounds)
using the different petrophysical models. The uncertainty bounds correspond to ±1
standard deviation. The vertical lines show the true total water volume. (b) The
corresponding changes in the amount of moisture within the extracted volume
relative to baseline. The vertical lines show the true change in total water volume.
(c) The contribution of different variables to the variance of total moisture of each
petrophysical models. (d) Additional variance (i.e., uncertainty) caused by
List of Figures
xv
uncertain porosity values (0.32±0.032). The contribution from uncertain porosity is
significant in most cases, especially when the variance in saturation is low. ....... 228
Figure 7 (a) Electrical resistivity tomography estimated changes in volume of water in
four selected cells. The vertical lines indicate the true change. (b) Scatter plots
showing the fit for change in volume of water at individual cells using the 15 Archie
models. The red dashed line in each plot is the best‐fit line of the scatter points. 230
Chapter 7
Figure 1 A diagram summarizing the major findings in this thesis and their relation to
the ERT workflow. ......................................................................................................... 250
Figure 2 (Top) Parameter space of a 3-parameter layered ERT problem using 24 surface
electrodes in dipole-dipole configuration. The axis represents the uniform
parameter value of each layer in log scale. The true resistivity for all three layers are
100 𝛀 𝐦. The red cross indicate the true parameter value. The data misfit surface
(left) and the derived streamlines to the data misfit mimima (right) are plotted.
(bottom) Gradient fields using 500 samples. The red polygons are the loops and the
black cross is the true values. The gradient field somewhat point towards the true
minima. 9 loops are identified, spanning a large fraction of the parameter space.
.......................................................................................................................................... 254
List of Tables
xvi
List of Tables
Chapter 3
Table 1 Table of error models reported in the literature ................................................ 125
Chapter 4
Table 1 “True” coupled hydrogeophysical model parameters used for synthetic
experiments. It is developed based on the Sellafield field trial. *Only parameters for
the main zone are listed below. #Leak location for some cases is (33.4534, -14.4303)
instead. Note that for all cases the leak location is at the water table. .................... 181
Table 2 Summary of synthetic cases. All cases converge in seven iterations. ............. 182
Table 3 Baseline coupled hydrogeophysical model parameters used for the parameter
estimation from the Hatfield field ERT data. *The domain consists of 3 meters of top
soil and a uniform main zone. Only parameters of the main zone are listed below.
........................................................................................................................................... 189
Table 4 Summary of cases for the Hatfield field example ............................................. 191
Table 5 Global sensitivity analysis results using the Morris (1991) method on selected
parameters on the Hatfield coupled hydrogeophysical model. The parameter
ranges considered and the mean absolute elementary effect (|EE|) are reported.
Parameter value combinations from ten realizations with the lowest RMSE are also
reported. .......................................................................................................................... 195
Chapter 5
Table 1 Parameters used for the water injection experiment. ....................................... 220
List of Acronyms
xvii
List of Acronyms
CO2 Carbon dioxide
DC Direct current
EM or EMI Electromagnetic induction
ER Environmental remediation
ERT or ERI Electrical resistivity tomography or Electrical resistivity imaging
GPR Ground penetrating radar
HLW High-level radioactive waste
IFRC Integrated Field-Scale Subsurface Research Challenge
IP Induced polarisation
McMC Markov chain Monte Carlo
NRZ Naturally- reduced zone
PRA Probabilistic risk assessment
QoI Quantity of interest
TCE trichloroethylene
TDEM Time domain EM
VOF Value of flexibility
VOI Value of information
Introduction
18
1. Introduction
1.1 Background
Effective characterisation is essential to successful management of
environmental sites (Artiola et al., 2004). They are, however, inherently labour- and
cost-intensive because point samples are needed to be obtained from boreholes. It is
often difficult to piece together the governing processes at the site based on the
individual point samples. These challenges are exaggerated in nuclear sites where site
access is restricted and risk of exposure of contamination is above average.
Geophysical methods have been used in the last two decades to improve the
effectiveness of site characterisation because they can “scan” the subsurface rapidly
like hospital scanners do (Binley et al., 2015). Therefore, they can provide information
about subsurface conditions at a spatial and temporal resolution that is not attainable
by point measurements (French et al., 2014). Like hospital scanners, geophysical
methods do not directly detect the quantity of interest (QoI) (e.g. pore water solute
concentration). Therefore, we need to understand how geophysical responses are
linked to the QoI. In order to make geophysics more useful for nuclear site
characterisation, we also need to understand how to better extract site information
from geophysical data, how errors and uncertainties propagate, and how to more
closely tie geophysical data to site conceptualization.
1.2 Objectives and aims
Using electrical resistivity tomography (ERT) (Binley, 2015a; Daily et al., 2005)
as an example, the primary objective of this work is to develop methods to better
quantify and improve the amount of information ERT can provide to aid site
characterisation. Electrical resistivity is related to subsurface material and fluid
properties; yet this relationship is controlled by multiple material and rock properties
and is uncertain. Moreover, all the data collection and interpretation stages in ERT
propagates through the interpretation workflow and contribute to the uncertainty of
the final interpretation of the ERT data to infer the quantity of interest (QoI), whether
it is soil water content, hydraulic parameters, or parameters describing the leakage of
Introduction
19
a potential contaminant from a storage facility. Prior to this study, little work had been
undertaken to specifically address the potential issues with uncertainty in using ERT
for hydrological predictions. Among the few work that attempted to address them, the
focus has been on either improving the inversion method, or reducing the uncertainty
in Bayesian model selection. None of them has undertaken a whole-system approach
for uncertainty quantification, nor have they evaluated aspects of uncertainty
propagation that does not depend on the choice of inversion methods (e.g.
measurement errors, petrophysical relationships).
The specific aims identified to fulfil the above were to:
Identify the sources of uncertainty in ERT data collection, modelling, inversion,
and interpretation
Assess the statistical distribution and correlation of ERT measurement errors
Assess the benefit of leak parameter estimation using ERT data directly (i.e.
without inversion)
Examine the extent to which uncertain petrophysical relationships affect the
estimation of soil water content (and its temporal changes)
To achieve the above aims the project objectives were to:
Review the use of geophysical data for nuclear site characterisation worldwide
(chapter 2)
Introduce the sources of uncertainty in ERT (chapter 3)
Conduct statistical analysis and develop a new model on ERT measurement errors
(chapter 4)
Develop a coupled hydrogeophysical data assimilation approach for leak
parameter estimation without reliance of ERT images (chapter 5)
Observe variability of petrophysical relationships in field soil samples, use it to
populate a range of petrophysical models, and examine the variability in the
estimated moisture content maps when ERT data is subjected the different
petrophysical models (chapter 6)
Introduction
20
We have traced the normal workflow for geophysical studies and consider it as
a pipeline for the propagation of both information and uncertainty and it can be used
to illustrate the relationship between the chapters (Figure 1). The chapters correspond
to three areas of interest for further investigation. An ERT study begins with designing
the survey and then collecting the measurements. Then the data collected is inverted
to obtain images of geophysical properties. The images are interpreted to understand
the cause of the behaviour observed in the images. Finally, such interpretation may be
applied for prediction of future events.
Figure 1 The ERT workflow showing the various stages of conducting ERT survey and analysing its
data. It also serves as a pipeline where information and uncertainty is propagated along. The
annotation shows the relation between the chapters in this thesis and the workflow.
Funded in part by the Nuclear Decommissioning Authority (NDA), this project
has a focus on leak detection using ERT at nuclear sites. However, it was thought that
some of the findings and conclusions from this study would improve the general
understanding of using ERT for site characterisation and monitoring, and would be
applicable for the deployment of ERT at other sites. It was also thought that some of
the findings are applicable to the use of other geophysical methods for site
characterisation.
1.3 Outline
Chapter 2 and 3 includes two literature review on the subject matter discussed
in the thesis. The first one provides details on the context of the application of near-
Introduction
21
surface geophysical methods for nuclear site characterisation. The second review
provides a detail discussion on the various sources of uncertainties in ERT by
summarizing exisiting research and identifying knowledge gaps.
Chapter 4 (Tso et al., 2017) describes an analysis of ERT measurement errors
using permanently installed ERT arrays. Various types and formulation of
measurement errors are assessed. A new model for ERT measurement errors is
proposed to handle potential bias of faulty electrodes. This work is an essential first
step to handle uncdertaintly propagation from ERT data in the hydrogeophysics
workflow.
Chapter 5 describes a novel geoelectrical leak detection method using coupled
hydrogeophysical modelling and data assimilation techniques. The ERT data
corresponding to the leak is used to provide information of the leak parameters and
reduce uncertainty in the geological conceptualization of the site. This work highlights
that in previously characterized sites (as in most nuclear sites), geophysical data can
be a powerful tool to estimate leak parameters using a minimal amount of boreholes.
Chapter 6 (also as Tso et al., 2019) explores the utility of inversion-based
estimates of moisture content from ERT under the influence of uncertain petrophysical
relationships. Field data shows that even cores within the same unit can show
significant variation in petrophysical relationships and if the full range is considered,
moisture content estimates can be highly uncertain. We advocate for the improved
consideration of petrophysical uncertainty in future work.
The above is followed by the discussion summary and conclusion sections.
Part of this work has been conducted using the coupled hydrogeophysical code
PFLOTRAN-E4D (Johnson et al., 2017), as illustrated in Figure 2. It was part of the
PFLOTRAN software releases until the end of this PhD. An alternative approach to
perform coupled hydrogeophysical simulation by running PFLOTRAN and E4D
separately is outlined in the Appendix.
Introduction
22
Figure 2 Snapshot showcasing coupled hydrogeophysical modelling using PFLOTRAN-E4D. The 2D
flow and transport model simulates tracer injection at an injector in the upper left of the domain.
Groundwater movement is towards an extraction well to the lower right. An ERT imaging cell with 4
boreholes (20 electrodes each) is located at the centre of the domain. The animation shows that as the
conductive tracer migrate through the ERT imaging cell, there is a corresponding increase in electrical
conductivity in the ERT imaging cell.
Geophysical methods for nuclear site characterisation
23
2. Geophysical methods for nuclear site characterisation
Manuscript prepared for journal submission as Tso, C.-H.M., Kuras, O., Binley, A.
(201x) Geophysical methods for nuclear site characterisation
Background
24
Background
Site characterisation, in the context here, involves desktop analysis of historic
studies, making observations and collecting data in the field and interpreting them in
order to build up a conceptual understanding of the geology, hydrogeology,
hydrology, and contaminant transport processes at the site. This understanding allows
assessments of exposure pathways and provides justification for clean-up decisions at
the site. Geophysical methods allow mapping and monitoring subsurface properties
and processes at resolutions and coverages that would otherwise be impossible to
attain by other point-scale methods. The chapter summarizes the current applications
of geophysical methods at nuclear sites and identifies critical gaps to be addressed in
future work.
Improved understanding of a site’s surface and subsurface conditions can
greatly reduce the costs and risks of decommissioning. In particular, uncertainty in the
site conditions contributes to tremendous financial and safety risks for multi-decade,
multi-billion pound decommissioning projects. Conventional methods are cost- and
labour-intensive because, traditionally, they rely on invasive, numerous local-scale
measurements. The interpretation drawn from these methods may not represent the
site-scale behaviour of the contaminant transport process. The above underpins a
major discrepancy between the principle and implementation of contaminated land
legislation. For example, in the U.K. contaminated land law, source-pathway-receptor
linkage of the contaminant needs to be established in order to determine risk and
responsibility (Environmental Protection Act 1990 - Part IIA Contaminated Land:
statutory guidance). This has been proved to be very difficult to achieve in the
subsurface environment. Geophysical methods offer a promising alternative as they
provide much greater site coverage. Their short data collection cycle also allows time-
lapse monitoring of transport processes. Much of the development in inversion has
been focused on improving the resolution of estimates or joint inversion of different
data types. Our understanding of a site, however, is always subject to uncertainty and
complicated by inconsistent scales of measurements. More robust methods to reduce
uncertainty in interpreting different site data is needed to improve site characterisation.
Conventional Methods
25
Site characterisation on nuclear sites (and ultimately the need for geophysics)
can have very diverse drivers, often associated with different regulatory issues,
different environmental hazards, and also different funding streams. More
importantly, the different needs of characterisation are often related to different spatial
and temporal scales of the problem. At the same time, some needs arise from a
‘civil/nuclear engineering’ angle, while others from a more holistic ‘environmental
assurance’ perspective. For example, the recent Sellafield leak monitoring work fell
under ‘decontamination & decommissioning (D&D)’, which addresses immediate
risks and often precedes more strategic ‘site restoration’ and ‘environmental
remediation (ER)’ tasks (timescales = many decades). There is also the (separate) issue
of long-term geological disposal, which requires characterization of the deeper
subsurface. At Sellafield and other NDA sites, all near-surface issues fall under the
responsibility of a ‘Land Quality’ directorate, and UK regulatory oversight comes from
the Environment Agency (EA) and the Office for Nuclear Regulation (ONR).
In this review, some conventional methods used in nuclear site characterisation
are outlined. Then, the application of geophysics at a number of selected nuclear sites
is reviewed. Finally, problems and aspects missing in current geophysical approaches
are summarized, and suggestions made for possible solutions. This report only focuses
on near-surface site characterisation (e.g. the shallowest 200 m). Geophysics can also
be used in deep repositories, but they are used to address different needs. For a review
of the technological development in characterizing potential deep repositories, the
reader is referred to the work of Tsang et al. (2015).
Conventional Methods
Nuclear site characterisation requires careful planning. Desk studies and site
walk-overs allow building up of knowledge of the site’s history and current condition.
They include surface mapping of geology, studying previous site records and
investigation reports, and obtaining information from nationally held databases
(Bayliss and Langley, 2003). They can provide the first evidence of site condition and
can help determine the feasibility of a field study proposal. Once preliminary site
Conventional Methods
26
conditions are understood, and health, safety, and logistical issues are resolved,
intrusive surveys can be arranged.
Typically, conventional characterisation of nuclear sites involves the drilling of
boreholes, pitting and logging of rocks, single- and cross-hole hydraulic testing, and
monitoring of water levels (Bayliss and Langley, 2003). Groundwater sampling is often
considered as the most important aspect of contaminated land assessment since they
provide direct evidence of the radioactive level and the presence of other chemical
constituents in groundwater.
Tracer methods have also been used extensively to evaluate the contaminant
transport behaviour at sites, for example, the US sites at Hanford (Ahlstrom et al., 1977)
and Savannah River (Webster et al., 1970). They are often used to infer solute advection
and dispersion properties of a site, which controls the rate at which a contaminant
plume spreads. Stochastic methods (e.g. Gelhar and Axness, 1983; Zhang and Neuman,
1990) can be used to relate small-scale tracer test results to large-scale displacements of
contaminants.
Geophysical surveys, such as surface and cross-borehole seismic and surface
electromagnetic (EM), are common practice in conventional site characterisation.
Common methods include electromagnetic conductivity mapping and sounding,
resistivity profiling, and ground penetrating radar. The conventional use of geophysics
distinguishes itself from its successors, not in the measurement techniques but its role
in the investigation. Geophysics was thought to only provide a secondary, indirect
means of characterizing a site prior to or in conjunction with intrusive work (Bayliss
and Langley, 2003). Likewise, down-hole geophysical logging is also often used during
the installation of boreholes. Again, conventional site characterisation does not
consider it as a part of the formal investigation. For example, natural gamma logs are
seen to assist overall approximation of hydraulic conductivity and allow data to be
cross-correlated with other field data (Bayliss and Langley, 2003). In recent
applications, however, both borehole and surface geophysics play a much more
important role in nuclear site characterisation. A detailed discussion is provided in the
next section.
Geophysical Methods
27
As an illustration, all of the conventional characterisation tasks mentioned
above were completed for part of the Sellafield site before 2000 (Bowden et al., 1998).
The site was one of the two potential areas (the other being Dounreay, Scotland) to
construct a low- and intermediate-level waste repository (Norton et al., 1997). Their
results are reported respectively in thirty-some NIREX reports published between 1993
and 1998.
Geophysical Methods
The use of geophysics for site characterization in nuclear sites can be traced back
to the 1970s (e.g. Edwards, 1977; Robins, 1979). However, its potential use for
characterising and monitoring nuclear sites was not recognized until the 1980s
(Morrison et al., 1987). An idealized 2-D nuclear waste repository was used to show
the effectiveness of surface-borehole ERT (Asch and Morrison, 1989). This review uses
two U.S. Department of Energy legacy sites—the Hanford site and the Savannah River
Site as case studies to discuss the development of applications of geophysical methods
in nuclear sites. This review also covers sites from the U.S. Department of Energy's
(DOE's) Integrated Field-Scale Subsurface Research Challenge (IFRC), a new program
that commits multi-investigator teams to perform large, benchmark-type experiments
on formidable field-scale science issues. IFRC consists of three legacy sites: Hanford
300 areas, Washington, Rifle, Colorado, and Oak Ridge, Tennessee. Finally, a summary
of applications in the United Kingdom and other countries are provided.
Hanford Site and Hanford 300 Area IFRC Site
The Hanford Site is located in south-central Washington State, the United
States. It was selected in 1943 as part of the secretive project to build an atomic weapon
(the Manhattan project) and was later expanded to produce weapon-grade and fuel-
grade plutonium. From 1945 to 1986, Hanford produced 65% of the plutonium
produced in US government-owned reactors (67 metric tons) and reprocessed 96,900
metric tons of Uranium (Gephart, 2010, 2003). It is among the largest open sites which
the US Department of Energy is obligated to clean-up according to US environmental
Geophysical Methods
28
laws. A more detailed review of the near-surface work done at Hanford is provided by
Johnson et al. (2015a).
The Hanford Site is built along a section of the Columbia River and is divided
into the 100, 200, and 300 Areas. The 300 Area was responsible for producing uranium
fuel rods. The fuel rods were then sent to the five chemical separation plants in the 200
Area to extract plutonium. Each of the separation and extraction processes used
complex, toxic, and corrosive chemicals that ultimately produced large amounts of
high-level radioactive waste (HLW), though the process has improved significantly
over the years. The extracted plutonium was sent to the nine plutonium production
reactors in the 100 Area. Currently, the 200 Area stores most of the site’s legacy waste
facilities.
The 300 Area is located at the east reach of Columbia River, while the 100 Area
aligns with the River’s north reach, and the 200 Area is located in Hanford’s Central
Plateau. Figure 1 shows a map of the Hanford site. The geology of Hanford mainly
consists of two formations: (1) the upper, Hanford Formation hosting the unconfined
aquifer in which groundwater flows; (2) the underlying, semi-confining Ringold
Formation. The interface between the permeable Hanford Formation and the relatively
impermeable Ringold Formation is a critical hydrogeological contact controlling the
vertical flow and transport of contaminated groundwater into the Columbia River
(Mwakanyamale et al., 2012).
The use of electrical potential variations to detect leaks at Hanford can be traced
back to the 1970s (Key, 1977). At the turn of the century, virtually all types of surface-
based geophysical methods have been tested at Hanford, including electrical resistance
tomography (ERT), ground-penetrating radar (GPR), numerous electromagnetic,
magnetic, seismic, and gravity methods. Over 250 geophysical surveys have been
conducted in portions of every “Area” of the Hanford Site (Last and Horton, 2000).
Geophysical Methods
29
Figure 1 Aerial view of the Hanford Site (Johnson et al., 2015a)
The 200 Area is located away from the river and its major concern is the
potential that contaminated water reaches regional groundwater or discharge to the
adjacent Columbia River. A basalt unit underlying the Ringold Formation can be seen
as a flow barrier yet its extent is unknown. Williams et al. (2012a, 2012b, 2012c, 2012d)
combined seismic reflection, vertical seismic profile, geologic cross-section, and well
control data to infer the subsurface basalt topography and showed that there is a
significant gap known as the Gable Gap in the basalt flow barrier (up to a few km
across), which provides a possible flow path for contaminants originating from the 200
Area.
Closer to the land surface, there are 40 single-shell HLW storage tanks in the B-
complex (the oldest one in Hanford) of the 200 Area, many of which have leaked or
experienced overfill episodes, and several outlying subsurface infiltration galleries.
The waste streams introduced to the vadose zone were highly saline and created zones
of elevated electrical conductivity. The waste poses a significant risk to groundwater
quality, and determining the distribution of vadose zone contamination remains one
of the most significant challenges limiting remediation and closure of Hanford Site
Geophysical Methods
30
waste disposal facilities. ERT surveys were conducted to determine the distribution of
highly conductive waste in the subsurface. Electrodes were laid parallel and
perpendicular to the tanks, trenches, and cribs (Rucker et al., 2007). A comparison
between the tank release history and groundwater sampling shows that higher
disposal volumes were generally accompanied by more dilute waste. One exception is
at the BY-Cribs Area, which was subjected to high waste volumes with high ionic
strengths. The inverted results are consistent with the contaminant release history
(Johnson and Wellman, 2013). They identify a main plume for the BY-Cribs Area which
is oriented directly beneath the cribs (i.e. unlined underground box-like structures to
dispose of liquid contaminants (Gephart, 2003)) have not only migrated vertically but
also laterally. An eastward trending lobe that dips downward at ~30o appears to
originate from the northeastern-most crib and a southeast trending lobe that appears
to be an extension of the main plume beneath the cribs. Figure 2 shows the estimated
plume for the BY-Cribs Area.
Figure 2 Estimated conductivity at the Hanford BY-Cribs (Johnson et al., 2010; Johnson and Wellman,
2013)
Within a tank farm there is typically extensive metallic infrastructure including
pipes, tanks, wells, electrical lines, distribution manifolds, and fences that complicates
the use of geophysical methods that aim to estimate electrical properties in the
subsurface. For example, previous studies in the 200 Area show poor ERT resolution
within the tank farm (Johnson and Wellman, 2013). Given the strong influence of
metallic infrastructure on electrical resistivity response, it may be advantageous to
Geophysical Methods
31
incorporate the infrastructure directly into the acquisition and modelling approaches
(Johnson et al., 2014). The steel-cased wells originally used for drywell geophysical
logging, for example, can be used as long electrodes in an electrical resistivity survey—
the concept was first tested by Ramirez (Ramirez et al., 1996). Recently, the method is
applied using a pilot-scale field experiment (Rucker, 2012) and within an actual tank
farm to image suspected releases from a tank (Rucker et al., 2013).
Ramirez et al. (1996) compared the pre-spilt and after-spilt ERT images of a
contaminant plume as it develops in soil under a tank already contaminated by
previous leakage. They concluded the new contaminant plume can still be detected
without the benefit of background data. Binley et al. (1997) tested the method of
electrical current imaging for leak detection at a mock tank set up in the 200 East Area
in Hanford. Both leaks from the centre of the tank and from the side were considered.
Unlike ERT, electrical current imaging aims to resolve the distribution of the
percentages of applied current within the electrode array and it can be sensitive to
small leaks. Daily et al. (2004) pioneered the near-real-time and remote monitoring of
leaks in underground storage tanks using ERT at the Hanford 200 East Area, as part of
a 5-party “mock tank” demonstration study for geophysical leak detection
technologies (Barnett et al., 2003, 2002). Based on the ERT images, a “leak/no-leak”
decisions is made daily. The ‘blind test’ yielded a 57% success rate, which is further
improved to 87% after defining a new criteria during re-analysis. The single-shell tank
used is characteristic of the 177 of them at Hanford and another 51 at Savannah River
currently containing waste. A number of steel-cased wells adjacent to the tank are used
as ‘long’ electrodes to sense bulk resistivity beneath the tank as well as depth-averaged
images of resistivity variation. 54,000 litres of sodium thiosulfate were episodically
released from a steel tank in a blind test lasting 110 days. Each day during the test a
leak or no-leak condition was declared based solely on analysis of the electrical data.
The summary diagnostic measure proposed therein, the mean logarithmic ratio (MLR),
is later used for value of information (VOI) studies to evaluate the benefit of using
different ERT array to monitor possible leakage from geological storage of carbon
dioxide (Trainor-Guitton et al., 2013b). The dataset from this study (Daily et al., 2004)
is later used for the first stochastic Markov-chain Monte Carlo (McMC) inversion of
Geophysical Methods
32
time-lapse ERT data (Ramirez et al., 2005). This technique combines prior information,
measured data, and forward models to produce subsurface resistivity distribution that
is most consistent with all available data. It allows quantification of the uncertainty of
a generated estimate, and allows alternative model estimates to be identified,
compared, ranked, and rejected. Similar leak detection systems are also used in later
studies at the site (Calendine et al., 2011; Glaser et al., 2008). The above is a good
example to illustrate how nuclear applications can set trends in geophysical
methodology.
The US Department of Energy (DOE) also initiated a treatability test program
to desiccate a portion of the vadose zone to minimize migration of contaminants
towards the water table, usually achieved by injection of non-reactive gas. The promise
of the technology relies on reducing the moisture content of the vadose zone, and
therefore monitoring its evolution over time is essential. A field test of using time-lapse
ERT to monitor desiccation of the 100-m thick vadose zone at the Hanford 200 Area
BC-Cribs and Trenches was conducted. The geophysical results show the creation of a
desiccation plume and map its evolution over time (Truex et al., 2013a). The moisture
content at the final time step agrees generally with independent neutron logging
measurements. Neutron logging and other methods, however, cannot map the change
in moisture content before and after the gas injection.
Finally, geophysical methods have also been used to assist the engineering
aspect of the remediation effort at the Hanford 200 Area. Specifically, they can be used
to map pipelines in the subsurface. Since high-density GPR over a large area is not
economically feasible nor necessary, electromagnetic induction (EM) and magnetic
gradiometry surveys were conducted by towing the equipment behind an all-terrain
vehicle outside TX and TY tank farms, covering ~40 hectares (Myers et al., 2008a,
2008b). The identification of pipes in the subsurface can help minimize false-positive
interpretation related to a potential contaminant disposed to the ground.
The Hanford 300 Area contains infiltration ponds that were used to dispose of
waste from uranium fuel rods production. It is adjacent to part of the Columbia River,
Geophysical Methods
33
the largest river in the US by volume. There exist many uranium hot spots within the
groundwater that vary with seasonal fluctuations in Columbia River stage levels
(Williams et al., 2008). The highest desorption of uranium occurs when the river stage
is as high as the smear zone beneath former ponds and trenches. As stage decreases,
groundwater flow moves towards the river and carries the desorbed uranium.
One of the challenges at the Hanford 300 Area is to understand the potential
connectivity of contaminant plume to the Columbia River. Slater et al. (2010) used
waterborne electrical resistivity and induced polarization surveys to map the thickness
of the Hanford unconfined aquifer at the river bed, which is indicative of paleochannel
structures that cause preferential groundwater flow. Using the temperature difference
between groundwater and river water as a proxy, they also used fibre-optic distributed
temperature sensing methods to map groundwater discharge along a 1.5 km stretch of
shoreline adjacent to the 300 Area. Their results reveal variable, time-dependent, stage-
driven groundwater discharge and river water intrusion along the shoreline.
Mwakanyamale et al. (2012) conducted further resistivity and IP surveys using seven
profiles running approximately parallel to the Columbia River and ~20m apart to
estimate the elevation of the Hanford-Ringold contact connecting the aquifer and the
Columbia River.
Johnson et al. (2012b) installed a 3-D electrode array near the 300 Area bank to
monitor stage-driven river water intrusion and retreat using 4-D ERT. In particular,
they capitalize on the distinctive contrast between groundwater and river water
conductance. Methods of distilling data such as time-series and time-frequency
analysis help improve the results. Correlation analysis between stage and bulk
conductivity at several depths validates the location of a dominant groundwater/river
water exchange zone along the shoreline previously identified by Slater et al. (2010).
River water is commonly detected 250 m inland to the IFRC well field during
spring runoff peak flow. Three 2-D ERT lines were deployed to monitor intrusion
during the 2011 peak flow event. The moving water table boundary becomes
problematic when smoothness-constrained inversion is used. Wallin et al. (2013) show
Geophysical Methods
34
that removing regularization constraints between neighbouring elements near the
water table boundary can effectively enable the inversion to map sharp contrast at the
water table boundary. Using ERT and borehole electrical conductivity (EC) data
collected from the IFRC well field in 2008, Johnson et al. (2012b) devised a new
constrained inversion method that allows incorporation of known geostatistical,
discontinuous boundary, and known conductivity constraints. The results are
compared to borehole flowmeter data. Building upon these efforts, Johnson et al.
(2015b) devised a warping mesh inversion to monitor the 2013 spring runoff at the
IFRC well field. The computational mesh warps to the known water table boundary
without changing mesh topology, thereby facilitating consistent regularization
constraints in the time dimension. The inversion is highly effective that it is conducted
near-real-time via wireless transfer. The real-time imaging provided information
concerning the onset of river water intrusion to scientists conducting time-sensitive
microbial sampling at the site.
Kowalsky et al. (2005) described a method to jointly use borehole time-lapse
GPR traveltimes data and neutron probe data to estimate unsaturated hydraulic
properties fields, which they applied it to an infiltration experiment at the Hanford 200
Area ‘Sisson and Lu site’. Their method has jointly estimated petrophysical
parameters and it corrects for initial over-prediction near the edge of the water plume.
Their work is an excellent example of how the combined use of geophysics and point-
based monitoring data can improve both coverage and accuracy of the estimates of
field-scale soil hydraulic parameters and the related moisture distribution.
Lastly, the reader has referred also to a few other joint inversion studies at the
Hanford 300 Area, which may be extended to include geophysical data in the future.
For example, Murakami et al. (2010) assimilated large-scale constant-rate injection data
with small-scale borehole flowmeter data using a Bayesian geostatistical inversion
framework, the method of anchored distributions (MAD) (Rubin et al., 2010). Chen et
al. (2012) expanded the study of Murakami et al. (2010) by using MAD to assimilate
also results from two field-scale nonreactive tracer tests. Using the same dataset, Chen
Geophysical Methods
35
et al. (2013) compare several ensemble-based data assimilation techniques for aquifer
characterisation, which was the first study of its kind using field data.
Finally, data management and visualization have become an important topic in
nuclear site management and characterization. The Pacific Northwest National
Laboratory has recently launched an online platform called SOCRATES
(https://socrates.pnnl.gov/) (Truex, 2018) , where all well logs, remote sensing images
and historical groundwater monitoring data of the Hanford site is available for plotting
and download on a user-friendly browser application (Figure 3). It includes options to
visualize selected wells only and several useful tools to visualize plume evolution over
time. Of relevance to geophysics is that the SOCRATES platform also includes near-
real-time ERT inversion images from the Hanford permanent monitoring network.
Such integrated data management platform can be valuable for deciding the
appropriate remediation strategy.
Figure 3 A screenshot for the browser-based integrated data platform SOCRATES for the Hanford site.
It serves as a centralized portal for all data collected at Hanford Site. It includes a wide variety of tools,
such as those for filtering data and exporting data and model domain for flow and transport modelling.
Savannah River Site
Most of the plutonium not produced at Hanford is produced at Savannah River
Site near Aiken, South Carolina. The main contaminants of concern, however, are
Geophysical Methods
36
chlorinated solvents such as trichloroethylene (TCE). One of the first time-lapse
electrical resistivity tomography studies was conducted at the site by Daily and
Ramirez (1995), to test the monitorability of methane air sparging for in-situ enhanced
bioremediation of TCE. They used a differential tomography method to remove all
static features of formation resistivity and found that the flow paths are confined to a
complex three-dimensional network of channels. Another series of tests obtained
images of water infiltration from the surface gives a similar conclusion, indicating the
flow regime is minimally modified by infiltration.
Subsequently, time-domain electromagnetic soundings were used to define the
electrical conductance of a clayey confining unit (aquitard), and shear-wave seismic
reflection was used to define the stratigraphic framework (Eddy-Dilek et al., 1997).
Later, three-dimensional (3-D) self-potential (SP) survey (Minsley et al., 2007) was
conducted at part of the site and the data was inverted using the model of Shi (1998)
to find an electrical current source model, taking into account the resistivity structure
derived from a 3-D spectral induced polarization survey at the same field location. The
sources and sinks of electrical current can be related to the zones of relatively high or
low redox potential and are therefore interpreted in the context of contaminated areas.
These results are reasonably correlated with contaminant concentration data obtained
from several ground-truth well measurements, indicating that the SP sources can be
an indicator of contaminated areas where electrochemical source mechanisms are
active.
Taking advantage of the often-coupled physical-geochemical-microbiological
properties of subsurface materials, Sassen et al. (2012) pioneered a reactive facies
concept and applied it to the site’s F-Area. The reactive facies concept is based on the
hypothesis that subsurface units exist with distinct distributions of coupled
physiochemical properties influencing reactive transport, such as effective surface area,
mineralogy, and hydraulic conductivity. Applying Bayesian data mining algorithms
to wellbore lithology, cone penetrometer testing, and cross-hole and surface seismic
data, Sassen et al. (2012) identified two distinctive reactive facies, which were
coincident to the two depositional facies at the site: a Barrier Beach and a Lagoonal
Geophysical Methods
37
facies. Wainwright et al. (2014a) expanded the method over large scales at high
resolution using a Bayesian hierarchal approach using recently acquired geochemical
data in addition to the dataset from Sassen et al. (2012). The facies estimates are then
used to obtain very high-resolution estimates of reactive facies-based transport
properties, such as percentages of fines, hydraulic conductivity, and Al:Fe ratio. The
modified approach is also a new formal methodology that quantifies the uncertainty
of reactive-facies properties, as well as the uncertainty in the spatial distribution of
reactive facies, using multiscale data sets.
Oak Ridge IFRC Site
At the Oak Ridge IFRC Site, Tennessee, the S-3 disposal ponds consisted of four
ponds built in 1951. They received a yearly volume of 7.6 million litres of acidic (pH<2)
liquid wastes consisting of nitric acid, uranium, technetium, cadmium, mercury, and
chlorinated solvents for 32 years. In 1983, the ponds were drained and filled to
neutralize the acidic wastewater, and they are covered with a multilayer cap to
minimize leaching. The meteoric water falling on the cap is drained and diverted to a
ditch surrounding the former S-3 ponds. The forced-gradient groundwater flow and
the infiltration of meteoric water create a mixing zone for the contaminants down-
gradient. The meteoric water has very different water chemistry than the plume. The
complexity of the geology at the site—saprolite overlaying fractured shale bedrock and
limestone interbeds – add challenge to the remediation of the site. Together,
understanding the subsurface heterogeneity within the plume pathways and how
recharge events affect the transport and attenuation of the plume are the major goals
of the geophysical investigation at this site.
The first geophysical studies conducted at the site were standard surface
resistivity tomography and seismic tomography along a survey line immediately
downgradient of the former S-3 ponds (Watson et al., 2005). Combined with limited
borehole data, they are used as a rapid and effective method for defining the location
of the high-ionic-strength plume and defining the transition zone between the saprolite
and bedrock so that a field plot can be established on-site to test microbially-mediated
reduction and immobilization of uranium (Wu et al., 2006).
Geophysical Methods
38
Subsequently, a Bayesian method was used to combine cross-hole seismic
travel times and borehole flowmeter test data to estimate hydrogeological zonation
(Chen et al., 2006). The Bayesian joint inversion approach permits information sharing
between the hydrogeological and geophysical data. The study identifies an extended
fracture zone along the primary axis of the plume. A later study inverts surface seismic
refraction data with depth constraints (Chen et al., 2010) using boreholes drilled for the
bioremediation pilot (Watson et al., 2005; Wu et al., 2006). They identify a zone of
persistent seismic slowness along two transects, confirming the presence of a
preferential flow path along the plume axis.
As more boreholes are drilled, more geochemical data are available. Joint
interpretation of geochemical and geophysical data appears to be a valuable tool to
study the hydrogeochemical processes in the heterogeneous subsurface at the site.
Kowalsky et al. (2011) tested 10 coupled inversion cases using 1-D or 2-D
hydrogeochemical models, combined with zero, one, or two ERT surveys from
electrodes installed along two boreholes. Gasperikova et al. (2012) used difference
inversion (Labrecque and Yang, 2001) to interpret data collected from 28 surface ERT
datasets collected along the same profile, and 22 cross-well datasets spanning 5
boreholes. The time-lapse inversion results show the evolution of nitrate levels
distribution in the subsurface, which is used to guide the site’s large-scale modelling
effort (Tang et al., 2010).
To better constrain geophysical inversion, detailed laboratory studies are
conducted to study the petrophysical properties of the contaminated and flushed
saprolites collected from the S-3 ponds (Revil et al., 2013a). A laboratory complex
conductivity study is also conducted on samples from the S-3 ponds (Revil et al.,
2013b), followed by inverting 15 time-lapse resistivity snapshots using the active time-
constrained approach (ATC) (Karaoulis et al., 2011b). The inversion results capture the
occurrence of an infiltration event during the winter of 2008–2009 with a dilution of
the pore water chemistry and an increase of the pH.
Geophysical Methods
39
Rifle IFRC Site
The Rifle, Colorado site is located on a floodplain adjacent to the Colorado
River. It contains three major hydrostratigraphic units: the fill layer, the shallow
unconfined aquifer Rifle Formation (~1-3 m below land surface), and the low-
permeability Wastach Formation (~5-8 m below land surface). The site was previously
used as a vanadium and uranium ore treatment facility. The Rifle site currently serves
as a community field laboratory for research in biogeochemical characterization,
bioremediation, subsurface microbial characterization and nutrient cycling. Recent
studies at the site report the presence of a naturally-reduced zone (NRZ) within the
aquifer sediments. The NRZ sediments had elevated concentrations of uranium,
organic matter, and geochemically reduced mineral phases, such as metal sulfides, and
were often associated with predominantly fine-grained sediment textures. The
conceptual model of NRZ formation is that roots, twigs, and other plant materials
accumulated during the river depositional process became buried and formed the
reduced sediments and abundance of iron-reducing Geobacteraceae communities.
Williams et al. (2009) conducted a spectral induced polarization survey to
resolve subsurface microbial activity at a high spatial resolution during bioremediation.
Fluids and sediments recovered from regions exhibiting an anomalous phase response
were enriched in Fe(II), dissolved sulfide, and cell-associated FeS nanoparticles. Flores-
Orozco et al. (2011) extended the previous dataset and collected time-lapse induced
polarization (IP) data during 100+ days of acetate injection, which was used to
stimulate microbial growth and hence to immobilize U(VI) in the unconfined aquifer.
The IP inversion identified zones with different redox characteristics, particularly an
increase in polarization effect accompanying the precipitation of iron sulphide. These
factors are critical in terms of controlling the rate and fate of bioremediation of uranium
in the subsurface. A later, more detailed inversion of the data finds that the
polarization effect of geochemically reduced, biostimulated sediment remains much
higher than background aquifer materials over a broad frequency bandwidth (0.06-120
Hz) (Flores-Orozco et al., 2013). Chen et al. (2013) combined the IP and aqueous
geochemistry time series using a hidden Markov model to estimate the timing of the
most probable transitions of redox states, which cannot otherwise be inferred from
Geophysical Methods
40
measurements directly. Wainwright et al. (2016) took a multi-step approach to map the
NRZ and the two hydrostratigraphic contacts at the site. They first inverted ERT and
time-domain IP (TDIP) data collected from ten of the surface lines. Statistical analysis
was then used to find correlations between the inverted geophysical model and
lithological logs information from 187 wells. Finally, the Bayesian reactive facies
approach (Sassen et al., 2012; Wainwright et al., 2014a) is used to map the most
probable surfaces of the two contacts and volume of the NRZ. This is the first study
that demonstrates the ability of TDIP imaging surveys for characterizing hotspots that
have unique distributions of subsurface lithological and biogeochemical properties. A
recent work by Chen et al. (2016) uses Bayesian methods to combine 47 years of
streamflow data upstream of the site and large-scale climate information to predict
future groundwater dynamics at the site. Results from this work can be used alongside
those from geophysical investigations to better understand subsurface structures and
processes at the site.
United Kingdom sites
Geophysical methods are also used extensively in sites in the United Kingdom
for nuclear site characterisation. Regrettably, many of work are not available in the
public domain. The use of electrical methods can be traced back to the 1970s (Robins,
1979) at Sellafield, Cumbria – U.K. ’s most complex nuclear site. Since then, there has
been a continuous effort to use different geophysical methods to map the subsurface
at different areas of Sellafield (Lean, 1998a, 1998b; Ross, 2004; Serco/Golder, 2010, 2009,
2008a, 2008b; TERRADAT (UK) LTD., 2012, 2004, 1998). Examples of methods used
include micro-gravity, downhole geophysical logging, EM61 (time domain metal
detector), GPR, and electrical resistivity imaging (ERI) surveys. In earlier years, gravity
and aeromagnetic data had been used (Kimbell, 1994).
The potential use of different geophysical methods is considered at all U.K.
nuclear sites (Booth, 1997), including the potential use of existing boreholes (Cooper,
1997). A comparative assessment among ten geophysical methods and case studies at
seven U.K. nuclear sites was also made (Cooper and Ross, 2005), where the authors
concluded that “the use of geophysical techniques at U.K. Nuclear Licensed Sites has
Geophysical Methods
41
a number of advantages over intrusive surveys in terms of dose rate to workers,
reduced time to obtain data, safety of workers and cost as intrusive surveys can be
prohibitively expensive on such sites”.
Most of the geophysical applications at U.K. nuclear-licensed sites take a rather
conventional approach. Geophysics plays only an assistive role in site characterisation
programs and its use focuses on identifying geophysical anomalies. There are a few
exceptions that explore the potential to monitor subsurface processes using borehole
electrical methods. At the Low-Level Waste Repository in Cumbria, the United
Kingdom, Kemna et al. (2004) has shown in a cross-hole ERT and IP survey that the
image of imaginary conductivity reveals additional information that allows
lithological decimation than the image of real conductivity. Throughout 2000, 2-D ERT
had also been used to monitor a tracer injection test at Drigg and the plume arrival
shows good agreement to background hydraulic gradient-based calculations (personal
communication with Andrew Binley, Lancaster University).
A full-scale field experiment applying 4D (3D time-lapse) cross-borehole ERT
to the monitoring of simulated subsurface leakage has been undertaken at a legacy
nuclear waste silo at the Sellafield Site (Kuras et al., 2016, 2015, 2014, 2011), similar to
the one at Hanford (Daily et al., 2004; Ramirez et al., 2005). A 4-D inversion (Kim et al.,
2009) was used to estimate the time-lapse changes in conductivity, and they have
revealed likely pathways of simulant flow in the vadose zone and upper groundwater
system. The geophysical evidence was found to be compatible with historic
contamination detected in permeable facies in borehole sediment cores, and with a
geological model based on wider scale borehole stratigraphy. The results suggest that
laterally discontinuous till units act as localized hydraulic barriers, which can
substantially affect flow patterns and contaminant transport in the shallow subsurface
at Sellafield.
In 2011, the Environment Agency reviewed the use of geophysical techniques
for nuclear site characterization (Environment Agency, 2011). The report ranks the
potential applicability of nine geophysical methods in nine different geological
environments in terms of providing information in six areas (e.g. geotechnical,
Geophysical Methods
42
hydrogeological, gas migration). The review’s focus was on the siting a geological
disposal facility (GDF). DC electrical resistivity method ranked low on the table,
probably because (i.) only a single-line surface survey is considered, (ii.) static but not
time-lapse surveys are considered, and (iii.) specific setups (such as leak detection or
tracer injection monitoring) is not considered. As already demonstrated in this chapter,
when used in the right context, DC electrical resistivity methods can be an extremely
powerful tool for nuclear site characterization.
Sites in other countries
For other countries, information about decommissioning of military-grade
nuclear facilities is often not available in the public domain. Previous work that is
related to disposal or storage of nuclear waste is briefly reviewed in the following.
In Russia, little is known about their application of geophysics for nuclear site
characterization. However, as an activity for a visit to Russia by a team of U.S.
Lawrence Berkeley National Laboratory scientists through the Russian-American
Center for Contaminant Transport, Frangos and Ter-Saakian (1996) performed
resistivity and IP surveys at the Chelyabinsk Nuclear Waste Site. They attribute the
anomalously low IP response recorded in some of the contaminated areas to a
radiolytic reaction with the dissolved nitrate, yielding oxygen which, in turn, reacts to
remove accessory pyrite from the host rocks.
At the French Institute of Radiological Protection and Nuclear Safety (IRSN)
experimental platform at Touremire (Aveyron, France), surface ERT was performed
for the underground research laboratory (for deep repository science) using two 2.5
km-long profiles to complement earlier 3-D seismic reflection investigations of
limestone and clay-rock formations (Gélis et al., 2010). Several distinct vertical low-
resistivity discontinuities were found, one associated with the regional Cernon fault,
while others are consistent with the location of well-identified secondary faults.
In northern European countries, geophysics is widely used for potential deep
waste repository characterisation, but not in near-surface remediation. Sweden is one
of the earliest countries to recognize the benefit of geophysical methods in nuclear site
Moving Forward
43
characterisation (Ahlbom et al., 1983). However, its work is mostly focused on deep
repositories (e.g. > 1 km). Note that Sweden will be the first country to house a deep
geological repository for its high-level radioactive waste (HLW). Borehole radar was
used in an early SKB project to monitor tracer transport in fractured crystalline rock
(Olsson et al., 1991; Olsson and Gale, 1995). Single-hole geophysics, as well as seismic
cross-hole and reflection experiments were also conducted in the late 1980s and 1990s
by the Swedish National Defense Institute and the Swedish Geological Co. Detailed
surface ground magnetic and resistivity measurements were taken for a potential deep
geologic repository for spent nuclear fuel in Oskarshamn (Stenberg, 2008). Similarly,
in Finland 30 borehole seismic surveys ware used to develop a complex site model
alongside existing magnetic and EM mapping data (e.g. Enescu et al., 2004).
In recent years, China has recognized the benefit of geophysical methods in the
design of disposal sites for its rapidly increasing number of nuclear power plants
(Zheng et al., 2000). Unfortunately, most of them are published in Chinese journals.
For example, a resistivity survey was conducted to select the disposal facility for the
Daya Bay Nuclear power plant near Hong Kong (Zhao, 2000). In Taiwan, geophysics
is also recognized as an important tool for nuclear site characterisation. For example,
geophysical methods were used to assist the building of the conceptual site model and
numerical transport models of the low-level nuclear waste repository in the western
Pacific island of Lan-Yu (Huang, 2013). Surface electromagnetic survey and high-
resolution audio-frequency magnetotelluric (AMT) surveys are used for the Xinchang
site in Beishan area in Gansu province, which is China’s proposed site for geological
disposal of high-level radioactive waste (Wang et al., 2018).
Moving Forward
The wide range of applications of geophysical methods in nuclear sites has
highlighted their utility and versatility in mapping subsurface facies and monitoring
time-lapse changes in water and contaminants in the subsurface. Geophysical methods
no longer only play an assistive role in identifying structural contrasts in the
subsurface; they have also emerged to be a highly flexible and cost-effective suite of
Moving Forward
44
tools to provide time-lapse monitoring of subsurface processes and high-resolution
imaging of the subsurface.
A necessary first step to quantify uncertainty in geophysical data is to
understand their errors. Like any measurements, basic research should be conducted
to determine the error levels, which is currently lacking. For example, ERT
measurements are generally taken using four electrodes, which may present some
unique challenges. Similarly, varying resolution in a geophysical image is well
recognized in the geophysical and hydrological community (e.g. Day-Lewis et al., 2005;
Singha and Moysey, 2006). Previous work has devised non-stationary calibration to
compensate such variation. It remains unclear, however, how to translate variations in
resolution to variations in uncertainty. When a geophysical property is translated to a
quantity of interest (QoI, e.g. solute concentration, saturation), a conversion using
petrophysical relationship needs to be assumed. The conversion is subject to
uncertainty and its impact on the estimates of the QoI and its uncertainty remain
unclear.
Nuclear site characterisation often spans decades and often lead to overlapping
work. Because of a large amount of historical information available, it is easy for
findings from previous characterization effort to be ignored. An effective, unifying
uncertainty and image appraisal framework from design, data collection, to
monitoring, is needed. It should take a sequential data collection approach that allows
evaluation of uncertainty reduction since there are alternating tasks of data collection
and interpretation. Data assimilation approaches can be used effectively to fuse
information from various site characterisation efforts together.
Currently, there is a lack of a framework to compare the reliability of different
characterisation options, particularly their ability to reduce risk and uncertainty.
Forward simulations can be used to establish the data reliability of different methods
(Bratvold and Begg, 2010; Nenna and Knight, 2013; Trainor-Guitton, 2014; Trainor-
Guitton et al., 2013b, 2013a, 2011). Decision analysis (Bratvold and Begg, 2010) and
probabilistic risk assessment (PRA) (Figure 4) (Paté-Cornell et al., 2010), which has a
Moving Forward
45
long history in evaluating nuclear reactor risks, can be used to rank different
alternatives.
Installation of field monitoring programs should be able to justify its costs. The
use of the value of information (VOI) (Eidsvik et al., 2015) to evaluate different
alternatives, whether it is to decide on installation or monitoring strategies, is common
practice in environmental applications. For example, they are used to decide the
optimal number of point samples to be collected (Back, 2007), worth of hydraulic
conductivity to determine maximum pumping in ecologically sensitive zones (Feyen
and Gorelick, 2005) or alternatives to groundwater remediation (Lee et al., 2012; Liu et
al., 2012). The VOI framework has also been applied to the use of geophysical methods
to solve subsurface problems under uncertainty, including determining the
monitorability of leak during geological storage of CO2 (Trainor-Guitton et al., 2013b),
placement of geothermal (Trainor-Guitton, 2014; Trainor-Guitton et al., 2014) wells,
and the value of using TDEM to monitor a groundwater desalinisation project (Nenna
et al., 2011). In CO2 storage, long-term monitoring programs need to be in place before
any data is collected. Some pioneering research on optimizing uncertainty reduction
(Eslick et al., 2014) and VOI analysis (Sato, 2011) can be found.
Although there are many merits of VOI calculations, they are context-specific.
Also, some characterization problems are more flexible than others to changes during
their implementation. A solution is to incorporate the value of flexibility (VOF) (Begg
et al., 2013) calculation when choosing characterisation alternatives. VOF has become
popular in manufacturing operation research and oil and gas markets. Programs that
can bring more side benefits to potential future uses and overall site objectives should
be favoured.
Uncertainty quantification or reduction creates no value in or of itself and does
not necessarily lead to better decision making (Bickel and Bratvold, 2008). Uncertainty
quantification methods must be decision-focused. Modelling details, including
uncertainty quantification, should only be included if it helps separate the alternatives
under consideration. Therefore, a decision should be made as soon as there is enough
information. An iterative decision analysis (Figure 4) is needed so that great detail is
Moving Forward
46
built into the model only when it is relevant to an important area. This approach also
accommodates learning and refinement, as well as a stopping rule that can reduce
work that does not increase value. On a related note, very often investigations are not
optimized to answer specific questions. The use of hypothesis-driven approaches issue
(Leube et al., 2012; Nowak et al., 2012) and task-driven approach for Bayesian
geostatistical design (Nowak et al., 2010) have emerged in recent work to address this.
A selection of other work simply focuses on designing programs to minimize risks (de
Barros et al., 2012; de Barros and Rubin, 2008; Tartakovsky, 2013; Varouchakis et al.,
2016).
Advances in data analytics allow real-time and continuous monitoring and
anomaly detection of data collected at nuclear site. For example, Schmidt et al. (2018)
used Kalman filter for real-time and continuous estimation of tritium and uranium
concentration from a live stream of specific conductance (SC) and pH data at Savannah
River Site F-Area. This is achieved by building data correlation between the input and
concentration data using principal component analysis (PCA) and coupling it with a
concentration decay model. The current implementation is limited to point sensors in
boreholes. Similar methods can be extended to incorporate geophysical data. If the
inverted images are needed, near-real-time inversions will need to be set up, like the
one that has been used to monitor and optimize remediation of contaminants real-time
at the Hanford site (Pacific Northwest National Laboratory, 2016).
In summary, there is a need to better understand the errors, uncertainties, and
information content of geophysical data and how they are translated and propagated
in the various stages of analysis. There is also a need to continue developing flexible
methods to incorporate different types of monitoring data from various
characterisation efforts. These efforts are pivotal to make geophysical data more useful
for nuclear site characterisation.
Figure 4 An illustration of the iterative decision analysis framework (Paté-Cornell et al., 2010).
References
47
References
Ahlbom, K., Carlsson, L., Olsson, O., 1983. Final disposal of spent nuclear fuel -
geological, hydrogeological and geophysical methods for site characterization.
Stockholm.
Ahlstrom, S.W., Foote, H.P., Arnett, R.C., Cole, C.R., Serne, R.J., 1977. Multicomponent
mass transport model: theory and numerical implementation (discrete-parcel-
random-walk version). Richland, Washington.
Asch, T., Morrison, H.F., 1989. Mapping and monitoring electrical resistivity with
surface and subsurface electrode arrays. Geophysics 54, 235–244.
Back, P.-E., 2007. A model for estimating the value of sampling programs and the
optimal number of samples for contaminated soil. Environ. Geol. 52, 573–585.
https://doi.org/10.1007/s00254-006-0488-6
Barnett, D.B., Gee, G.W., Sweeney, M.D., 2002. Results of Tank-Leak Detection
Demonstration Using Geophysical Techniques at the Hanford Mock Tank Site—
Fiscal Year 2001. Richland, Washington.
Barnett, D.B., Gee, G.W., Sweeney, M.D., Johnson, M.D., Medina, V.F., Khan, F., 2003.
Results of Performance Evaluation Testing of Electrical Leak-Detection Methods
at the Hanford Site Mock Tank – FY 2002-2003. Richland, Washington.
Bayliss, C.R., Langley, K.F., 2003. Nuclear Decommissioning, Waste Management, and
Environmental Site Remediation, Nuclear Decommissioning, Waste Management,
and Environmental Site Remediation. Elsevier. https://doi.org/10.1016/B978-
075067744-8/50027-8
Begg, S., Bratvold, R., Campbell, J., 2013. The value of flexibility in managing
Uncertainty in oil and gas investments, in: SPE Annual Technical Conference and
Exhibition. Society of Petroleum Engineers. https://doi.org/10.2118/77586-MS
Bickel, E.J., Bratvold, R.B., 2008. From uncertainty quantification to decision making in
the oil and gas industry. Energy, Explor. Exploit. 26, 311–325.
https://doi.org/10.1260/014459808787945344
Binley, A., Daily, W., Ramirez, A., 1997. Detecting leaks from environmental barriers
using electrical current imaging. J. Environ. Eng. Geophys. 2 (March), 11–19.
https://doi.org/10.4133/JEEG2.1.11
Booth, P.M., 1997. Sellafield site characterisation project: an appraisal of site
investigation techniques previously untrailed at BNFL’s sites. Risley, United
Kingdom.
Bowden, R.A., Bumpus, C., Littleboy, A.K., 1998. An overview and update of the site
References
48
characterization studies at Sellafield. Proc. Yorksh. Geol. Soc. 52, 125–137.
https://doi.org/10.1144/pygs.52.2.125
Bratvold, R.B., Begg, S., 2010. Making good decisions. Society of Petroleum Engineers,
Richardson, Texas, USA.
Calendine, S., Rucker, D.F., Fink, J.B., Levitt, M.T., 2011. Automated leak detection of
buried tanks using geophysical methods at the Hanford nuclear site, in: SAGEEP
2011. Annual Meeting of the Environmental and Engineering Geophysical Society.
Charleston, South Carolina.
Chen, J., Hubbard, S., Peterson, J., Williams, K., Fienen, M., Jardine, P., Watson, D.,
2006. Development of a joint hydrogeophysical inversion approach and
application to a contaminated fractured aquifer. Water Resour. Res. 42, 1–13.
https://doi.org/10.1029/2005WR004694
Chen, J., Hubbard, S.S., Williams, K.H., 2013. Data-driven approach to identify field-
scale biogeochemical transitions using geochemical and geophysical data and
hidden Markov models: Development and application at a uranium-
contaminated aquifer. Water Resour. Res. 49, 6412–6424.
https://doi.org/10.1002/wrcr.20524
Chen, J., Hubbard, S.S., Williams, K.H., Ficklin, D.L., 2016. Estimating groundwater
dynamics at a Colorado River floodplain site using historical hydrological data
and climate information. Water Resour. Res. n/a-n/a.
https://doi.org/10.1002/2015WR017777
Chen, J.S., Hubbard, S.S., Gaines, D., Korneev, V., Baker, G., Watson, D., 2010.
Stochastic estimation of aquifer geometry using seismic refraction data with
borehole depth constraints. Water Resour. Res. 46, 16. https://doi.org/W11539
10.1029/2009wr008715
Chen, X., Hammond, G.E., Murray, C.J., Rockhold, M.L., Vermeul, V.R., Zachara, J.M.,
2013. Application of ensemble-based data assimilation techniques for aquifer
characterization using tracer data at Hanford 300 area. Water Resour. Res. 49,
7064–7076. https://doi.org/10.1002/2012WR013285
Chen, X., Murakami, H., Hahn, M.S., Hammond, G.E., Rockhold, M.L., Zachara, J.M.,
Rubin, Y., 2012. Three-dimensional Bayesian geostatistical aquifer
characterization at the Hanford 300 Area using tracer test data. Water Resour. Res.
48, W06501. https://doi.org/10.1029/2011WR010675
Cooper, S., 1997. Assessment of the potential for downhole geophysics within the BHP
boreholes. Risley, United Kingdom.
Cooper, S., Ross, M.O., 2005. Application and comparative assessment of non-intrusive
site characterisation techniques. Daresbury, United Kingdom.
References
49
Daily, W., Ramirez, A., 1995. Electrical resistance tomography during in-situ
trichloroethylene remediation at the Savannah River Site. J. Appl. Geophys. 33,
239–249. https://doi.org/10.1016/0926-9851(95)90044-6
Daily, W., Ramirez, A., Binley, A., 2004. Remote monitoring of leaks in storage tanks
using electrical resistance tomography: Application at the Hanford Site. J. Environ.
Eng. Geophys. 9, 11–24. https://doi.org/10.4133/jeeg9.1.11
Day-Lewis, F.D., Singha, K., Binley, A.M., 2005. Applying petrophysical models to
radar travel time and electrical resistivity tomograms: Resolution-dependent
limitations. J. Geophys. Res. B Solid Earth 110, 1–17.
https://doi.org/10.1029/2004JB003569
de Barros, F.P.J., Ezzedine, S., Rubin, Y., 2012. Impact of hydrogeological data on
measures of uncertainty, site characterization and environmental performance
metrics. Adv. Water Resour. 36, 51–63.
https://doi.org/10.1016/j.advwatres.2011.05.004
de Barros, F.P.J., Rubin, Y., 2008. A risk-driven approach for subsurface site
characterization. Water Resour. Res. 44, 1–14.
https://doi.org/10.1029/2007WR006081
Eddy-Dilek, C.A., Looney, B.B., Hoekstra, P., Harthill, N., Blohm, M., Phillips, D.B.,
1997. Definition of a critical confining zone using surface geophysical methods.
Groundwater 35, 451–462.
Edwards, L.S., 1977. A modified pseudosection for resistivity and IP. Geophysics 42,
1020–1036. https://doi.org/10.1190/1.1440762
Eidsvik, J., Mukerji, T., Bhattacharjya, D., 2015. Value of information in the earth
sciences: Integrating spatial modeling and decision analysis. Cambridge
University Press, Cambridge.
Enescu, N., Ahokas, T., Heikkinen, E., Cosma, C., 2004. Integrated Geophysical
Characterization of a Hard Rock Site, in: EAGE 66th Conference & Exhibition.
Paris, France, pp. 7–10.
Environment Agency, 2011. Geophysical surveying techniques to characterise a site for
a geological disposal facility: A review of recent developments and NDA’s
proposals.
Eslick, J.C., Ng, B., Gao, Q., Tong, C.H., Sahinidis, N. V., Miller, D.C., 2014. A
framework for optimization and quantification of uncertainty and sensitivity for
developing carbon capture systems. Energy Procedia 63, 1055–1063.
https://doi.org/10.1016/j.egypro.2014.11.113
Feyen, L., Gorelick, S.M., 2005. Framework to evaluate the worth of hydraulic
References
50
conductivity data for optimal groundwater resources management in ecologically
sensitive areas. Water Resour. Res. 41, 1–13.
https://doi.org/10.1029/2003WR002901
Flores-Orozco, A., Williams, K.H., Kemna, A., 2013. Time-lapse spectral induced
polarization imaging of stimulated uranium bioremediation. Near Surf. Geophys.
11, 531–544. https://doi.org/10.3997/1873-0604.2013020
Flores-Orozco, A., Williams, K.H., Long, P.E., Hubbard, S.S., Kemna, A., 2011. Using
complex resistivity imaging to infer biogeochemical processes associated with
bioremediation of an uranium-contaminated aquifer. J. Geophys. Res.
Biogeosciences 116, 1–17. https://doi.org/10.1029/2010JG001591
Frangos, W., Ter-Saakian, S., 1996. Resistivity and induced polarization survey at a
Russian nuclear waste site. Berkeley, California, United States.
https://doi.org/10.1149/1.2221597
Gasperikova, E., Hubbard, S.S., Watson, D.B., Baker, G.S., Peterson, J.E., Kowalsky,
M.B., Smith, M., Brooks, S., 2012. Long-term electrical resistivity monitoring of
recharge-induced contaminant plume behavior. J. Contam. Hydrol. 142–143, 33–
49. https://doi.org/10.1016/j.jconhyd.2012.09.007
Gelhar, L., Axness, C., 1983. Three-dimensional stochastic analysis of macrodispersion
in aquifers. Water Resour. Res. 19, 161–180.
Gélis, C., Revil, A., Cushing, M.E., Jougnot, D., Lemeille, F., Cabrera, J., de Hoyos, A.,
Rocher, M., 2010. Potential of electrical resistivity tomography to detect fault
zones in limestone and argillaceous formations in the experimental platform of
Tournemire, France. Pure Appl. Geophys. 167, 1405–1418.
https://doi.org/10.1007/s00024-010-0097-x
Gephart, R.E., 2010. A short history of waste management at the Hanford Site. Phys.
Chem. Earth, Parts A/B/C 35, 298–306. https://doi.org/10.1016/J.PCE.2010.03.032
Gephart, R.E., 2003. A short history of Hanford waste generation, storage, and release.
Pnnl-13605 1–39.
Glaser, D.R., Fink, J.B., Levitt, M.T., Rucker, D.F., 2008. A summary of recent
geophysical investigations at the Department of Energy Hanford Nuclear Facility,
in: 2008 Chapman Conference on Biogeophysics. Portland, Maine.
Huang, H.-T., 2013. Radionuclide Migration in Low-level Radioactive Waste Disposal
at Lan-Yu Site (in Chinese). National Cheng Kung University.
Johnson, T., Triplett, M., Wellman, D., Northwest, P., 2014. Modeling of electrical
resistivity data in the presence of electrically conductive well casings and waste
storage tanks, in: WM2014 Conference, March 2 – 6. Phoenix, Arizona, USA, p. 7.
References
51
Johnson, T.C., Rucker, D.F., Glaser, D.R., 2015a. Near-Surface Geophysics at the
Hanford Nuclear Site, the United States, in: Treatise on Geophysics. Elsevier, pp.
571–595. https://doi.org/10.1016/B978-0-444-53802-4.00205-0
Johnson, T.C., Versteeg, R., Thomle, J., Hammond, G., Chen, X., Zachara, J., 2015b.
Four-dimensional electrical conductivity monitoring of stage-driven river water
intrusion: Accounting for water table effects using a transient mesh boundary and
conditional inversion constraints. Water Resour. Res. 51, 6177–6196.
https://doi.org/10.1002/2014WR016129
Johnson, T.C., Versteeg, R.J., Rockhold, M., Slater, L.D., Ntarlagiannis, D., Greenwood,
W.J., Zachara, J., 2012. Characterization of a contaminated wellfield using 3D
electrical resistivity tomography implemented with geostatistical, discontinuous
boundary, and known conductivity constraints. Geophysics 77, EN85.
https://doi.org/10.1190/geo2012-0121.1
Johnson, T.C., Versteeg, R.J., Ward, A., Day-Lewis, F.D., Revil, A., 2010. Improved
hydrogeophysical characterization and monitoring through parallel modeling
and inversion of time-domain resistivity and induced-polarization data.
Geophysics 75, WA27. https://doi.org/10.1190/1.3475513
Johnson, T.C., Wellman, D.M., 2013. Re-inversion of surface electrical resistivity
tomography data from the Hanford Site B-Complex.
Karaoulis, M., Revil, a., Werkema, D.D., Minsley, B.J., Woodruff, W.F., Kemna, a.,
2011. Time-lapse three-dimensional inversion of complex conductivity data using
an active time constrained (ATC) approach. Geophys. J. Int. 187, 237–251.
https://doi.org/10.1111/j.1365-246X.2011.05156.x
Kemna, A., Binley, A.M., Slater, L., 2004. Cross-borehole IP imaging for engineering
and environmental applications. Geophysics 69, 97–107.
https://doi.org/10.1190/1.1649379
Key, K.T., 1977. Nuclear waster tank and pipeline external leak detection systems.
Richland, Washington.
Kim, J.-H., Yi, M.-J., Park, S.-G., Kim, J.G., 2009. 4-D inversion of DC resistivity
monitoring data acquired over a dynamically changing earth model. J. Appl.
Geophys. 68, 522–532. https://doi.org/10.1016/j.jappgeo.2009.03.002
Kimbell, G.S., 1994. The interpretation of gravity and aeromagnetic data for the
geological investigation of the Sellafield area, west Cumbria. Proc. Yorksh. Geol.
Soc. 50, 103–112. https://doi.org/10.1144/pygs.50.1.103
Kowalsky, M.B., Finsterle, S., Peterson, J., Hubbard, S., Rubin, Y., Majer, E., Ward, A.,
Gee, G., 2005. Estimation of field-scale soil hydraulic and dielectric parameters
through joint inversion of GPR and hydrological data. Water Resour. Res. 41, n/a-
References
52
n/a. https://doi.org/10.1029/2005WR004237
Kowalsky, M.B., Gasperikova, E., Finsterle, S., Watson, D., Baker, G., Hubbard, S.S.,
2011. Coupled modeling of hydrogeochemical and electrical resistivity data for
exploring the impact of recharge on subsurface contamination. Water Resour. Res.
47, W02509. https://doi.org/10.1029/2009WR008947
Kuras, O., Wilkinson, P.B., Meldrum, P.I., Oxby, L.S., Uhlemann, S., Chambers, J.E.,
Binley, A., Graham, J., Smith, N.T., Atherton, N., 2016. Geoelectrical monitoring
of simulated subsurface leakage to support high-hazard nuclear
decommissioning at the Sellafield Site, UK. Sci. Total Environ. 566–567, 350–359.
https://doi.org/10.1016/j.scitotenv.2016.04.212
Kuras, O., Wilkinson, P.B., Meldrum, P.I., Oxby, L.S., Uhlemann, S., Chambers, J.E.,
Binley, A.M., Graham, J., Dewey, G., Atherton, N., 2014. A trial of 4D Cross-hole
electrical resistivity tomography (ERT) for detecting and monitoring subsurface
leakage and contaminant transport, supporting the decommissioning of legacy
silos at the Sellafield Site, UK, in: WM2014 Conference, March 2 – 6. Phoenix,
Arizona, USA.
Kuras, O., Wilkinson, P.B., Meldrum, P.I., Uhlemann, S., Chambers, J.E., 2015. Results
of the GEMS ERT trial and extended baseline measurements at Sellafield.
Keyworth.
Kuras, O., Wilkinson, P.B., White, J.C., Chambers, J.E., Meldrum, P.I., Ogilvy, R.D.,
2011. MSSS leak mitigation-leak detection phase 3: Desk study for ERT technology.
Keyworth.
Labrecque, D.J., Yang, X., 2001. Difference inversion of ERT data: a fast inversion
method for 3-D in situ monitoring. J. Environ. Eng. Geophys. 6, 83–89.
https://doi.org/dx.doi.org/10.4133/JEEG6.2.83
Last, G. V, Horton, D.G., 2000. Review of geophysical characterization methods used
at the Hanford Site.
Lean, C., 1998a. The use of surface geophysics in the geological characterisation of
Sellafield. Risley, United Kingdom.
Lean, C., 1998b. TerraDat geophysical site investigation at the buried channel area of
Sellafield. Risley, United Kingdom.
Lee, J., Liu, X., Kitanidis, P.K., Kim, U., Parker, J., Bloom, A., Lyon, R., 2012. Cost
optimization of DNAPL remediation at Dover Air Force Base Site. Groundw.
Monit. Remediat. 48–56. https://doi.org/10.1111/j1745
Leube, P.C., Geiges, A., Nowak, W., 2012. Bayesian assessment of the expected data
impact on prediction confidence in optimal sampling design. Water Resour. Res.
References
53
48, 1–16. https://doi.org/10.1029/2010WR010137
Liu, X., Lee, J., Kitanidis, P.K., Parker, J., Kim, U., 2012. Value of information as a
context-specific measure of uncertainty in groundwater remediation. Water
Resour. Manag. 26, 1513–1535. https://doi.org/10.1007/s11269-011-9970-3
Minsley, B.J., Sogade, J., Morgan, F.D., 2007. Three-dimensional self-potential
inversion for subsurface DNAPL contaminant detection at the Savannah River
Site, South Carolina. Water Resour. Res. 43, 1–13.
https://doi.org/10.1029/2005WR003996
Morrison, H.F., Beckers, A., Lee, K.H., 1987. Measurements of electrical conductivity
for characterizing and monitoring nuclear waste repositories. Berkeley, California,
United States.
Murakami, H., Chen, X., Hahn, M.S., Liu, Y., Rockhold, M.L., Vermeul, V.R., Zachara,
J.M., Rubin, Y., 2010. Bayesian approach for three-dimensional aquifer
characterization at the Hanford 300 Area. Hydrol. Earth Syst. Sci. 14, 1989–2001.
https://doi.org/10.5194/hess-14-1989-2010
Mwakanyamale, K., Slater, L., Binley, A., Ntarlagiannis, D., 2012. Lithologic imaging
using complex conductivity: Lessons learned from the Hanford 300 Area.
Geophysics 77, E397. https://doi.org/10.1190/geo2011-0407.1
Myers, D.A., Cubbage, B., Brauchla, R., O’Brien, G., Bergeron, M., Henderson, C., 2008a.
Surface geophyscial exploration of TX-TY tank farms at the Hanford Site: Results
of background characterization with ground penetrating radar. RPP-RPT-38104.
Richland, Washington.
Myers, D.A., Rucker, D.F., Levitt, M., Cubbage, B., Henderson, C., 2008b. Surface
geophyscial exploration of TX-TY tank farms at the Hanford Site: Results of
background characterization with magnetics and electromagnetics. RPP-RPT-
36893. Richland, Washington.
Nenna, V., Knight, R., 2013. Demonstration of a value of information metric to assess
the use of geophysical data for a groundwater application. Geophysics 79, E51–
E60. https://doi.org/10.1190/geo2012-0474.1
Nenna, V., Pidlisecky, A., Knight, R., 2011. Informed experimental design for electrical
resistivity imaging. Near Surf. Geophys. 9, 469–482. https://doi.org/10.3997/1873-
0604.2011027
Norton, M.G., Arthur, J.C.R., Dyer, K.J., 1997. Geophysical survey planning for the
Dounreay and Sellafield geological investigations. Geol. Soc. London, Eng. Geol.
Spec. Publ. 12, 335–343. https://doi.org/10.1144/GSL.ENG.1997.012.01.31
Nowak, W., de Barros, F.P.J., Rubin, Y., 2010. Bayesian geostatistical design: Task-
References
54
driven optimal site investigation when the geostatistical model is uncertain.
Water Resour. Res. 46, 1–17. https://doi.org/10.1029/2009WR008312
Nowak, W., Rubin, Y., de Barros, F.P.J., 2012. A hypothesis-driven approach to
optimize field campaigns. Water Resour. Res. 48, W06509.
https://doi.org/10.1029/2011WR011016
Olsson, O., Anderson, P., Gustafsson, E., 1991. PROJECT 9118 Site Characterization
and Validation - Monitoring of Saline Tracer Transport by Borehole Radar
Measurements, Final Report.pdf. Stockholm.
Olsson, O., Gale, J.E., 1995. Site assessment and characterization for high-level nuclear
waste disposal; results from the Stripa Project, Sweden. Q. J. Eng. Geol. 28, 17–30.
Pacific Northwest National Laboratory, 2016. An Award-Winning Subsurface
Software [WWW Document]. URL
https://energyenvironment.pnnl.gov/highlights/highlight.asp?id=2556 (accessed
9.16.19).
Paté-Cornell, E., Cochran, J.J., Cox, L.A., Keskinocak, P., Kharoufeh, J.P., Smith, J.C.,
Assessment, F.R., 2010. An introduction to probabilistic risk analysis for
engineered systems. Wiley Encycl. Oper. Res. Manag. Sci.
https://doi.org/10.1002/9780470400531.eorms0680
Ramirez, A., Daily, W., 1996. Detection of leaks in underground storage tanks using
electrical resistance methods:1996 results. Livermore, California.
Ramirez, A., Daily, W., Binley, A., LaBrecque, D., Roelant, D., 1996. Detection of Leaks
in Underground Storage Tanks Using Electrical Resistance Methods. J. Environ.
Eng. Geophys. 1, 189–203. https://doi.org/10.4133/JEEG1.3.189
Ramirez, A.L., Nitao, J.J., Hanley, W.G., Aines, R., Glaser, R.E., Sengupta, S.K., Dyer,
K.M., Hickling, T.L., Daily, W.D., 2005. Stochastic inversion of electrical resistivity
changes using a Markov Chain Monte Carlo approach. J. Geophys. Res. 110,
B02101. https://doi.org/10.1029/2004JB003449
Revil, A., Skold, M., Hubbard, S.S., Wu, Y., Watson, D.B., Karaoulis, M., 2013a.
Petrophysical properties of saprolites from the Oak Ridge Integrated Field
Research Challenge site, Tennessee. Geophysics 78, D21–D40.
https://doi.org/10.1190/geo2012-0176.1
Revil, A., Wu, Y., Karaoulis, M., Hubbard, S.S., Watson, D.B., Eppehimer, J.D., 2013b.
Geochemical and geophysical responses during the infiltration of fresh water into
the contaminated saprolite of the Oak Ridge Integrated Field Research Challenge
site, Tennessee. Water Resour. Res. 49, n/a-n/a. https://doi.org/10.1002/wrcr.20380
Robins, N.S., 1979. Electrical resistivity survey at the Windscale Works, Cumbria.
References
55
Oxfordshire, United Kingdom.
Ross, M., 2004. Consideration of available geophysical techniques for non-intrusive
survey of B291 trenches. Sellafield, United Kingdom.
Rubin, Y., Chen, X., Murakami, H., Hahn, M., 2010. A Bayesian approach for inverse
modeling, data assimilation, and conditional simulation of spatial random fields.
Water Resour. Res. 46, 1–23. https://doi.org/10.1029/2009WR008799
Rucker, D., Levitt, M., O’Brien, G., Henderson, C., 2007. Surface geophysical
exploration of B , BX , and BY Tank Farms at the Hanford Site : Results of
background characterization with ground penetrating radar.
Rucker, D.F., 2012. Enhanced resolution for long electrode ERT. Geophys. J. Int. 191,
101–111. https://doi.org/10.1111/j.1365-246X.2012.05643.x
Rucker, D.F., Myers, D.A., Cubbage, B., Levitt, M.T., Noonan, G.E., McNeill, M.,
Henderson, C., Lober, R.W., 2013. Surface geophysical exploration: Developing
noninvasive tools to monitor past leaks around Hanford’s tank farms. Environ.
Monit. Assess. 185, 995–1010. https://doi.org/10.1007/s10661-012-2609-x
Sassen, D.S., Hubbard, S.S., Bea, S.A., Chen, J., Spycher, N., Denham, M.E., 2012.
Reactive facies: An approach for parameterizing field-scale reactive transport
models using geophysical methods. Water Resour. Res. 48, 1–20.
https://doi.org/10.1029/2011WR011047
Sato, K., 2011. Value of information analysis for adequate monitoring of carbon dioxide
storage in geological reservoirs under uncertainty. Int. J. Greenh. Gas Control 5,
1294–1302. https://doi.org/10.1016/j.ijggc.2011.07.010
Schmidt, F., Wainwright, H.M., Faybishenko, B., Denham, M., Eddy-Dilek, C., 2018. In
Situ Monitoring of Groundwater Contamination Using the Kalman Filter.
Environ. Sci. Technol. 52, 7418–7425. https://doi.org/10.1021/acs.est.8b00017
Serco/Golder, 2010. Sellafield contaminated land and groundwater management
project – geophysical survey – micro-gravity, downhole logging, EM61, GPR and
ERI.
Serco/Golder, 2009. Sellafield contaminated land and groundwater management
project – Report for B291 Trench Area investigation.
Serco/Golder, 2008a. Sellafield contaminated land and groundwater management
project – geophysics survey – B291 Trench Area.
Serco/Golder, 2008b. Sellafield contaminated land and groundwater management
project – Geophysical data collection plan.
Shi, W., 1998. Advanced modeling and inversion techniques for three-dimensional
References
56
geoelectrical surveys. Massachusetts Institute of Technology.
Singha, K., Moysey, S., 2006. Accounting for spatially variable resolution in electrical
resistivity tomography through field-scale rock-physics relations. Geophysics 71,
A25. https://doi.org/10.1190/1.2209753
Slater, L.D., Ntarlagiannis, D., Day-Lewis, F.D., Mwakanyamale, K., Versteeg, R.J.,
Ward, A., Strickland, C., Johnson, C.D., Lane, J.W., 2010. Use of electrical imaging
and distributed temperature sensing methods to characterize surface water-
groundwater exchange regulating uranium transport at the Hanford 300 Area,
Washington. Water Resour. Res. 46, 1–13. https://doi.org/10.1029/2010WR009110
Stenberg, L., 2008. Detailed ground magnetic and resistivity measurements for a
potential deep geologic repository for spent nuclear fuel in Oskarshamn, in:
Ahlbom, K., Stephens, M. (Eds.), Investigations of Potential Repository Sites for
Spent Nuclear Fulel at Forsmark and Laxemar-Simpevarp, Sweden. SKB,
Stockholm, p. 43.
Tang, G., D’Azevedo, E.F., Zhang, F., Parker, J.C., Watson, D.B., Jardine, P.M., 2010.
Application of a hybrid MPI/OpenMP approach for parallel groundwater model
calibration using multi-core computers. Comput. Geosci. 36, 1451–1460.
https://doi.org/10.1016/j.cageo.2010.04.013
Tartakovsky, D.M., 2013. Assessment and management of risk in subsurface
hydrology: A review and perspective. Adv. Water Resour. 51, 247–260.
https://doi.org/10.1016/j.advwatres.2012.04.007
TERRADAT (UK) LTD., 2012. Geophysical survey to locate subsurface features and
services at B38, Sellafield, Cumbria. Cardiff, United Kingdom.
TERRADAT (UK) LTD., 2004. Geophysical investigations into the subsurface deposits
and structures. Cardiff, United Kingdom.
TERRADAT (UK) LTD., 1998. Geophysical survey trials carried out at Sellafield, West
Boundary. Cardiff, United Kingdom.
Trainor-Guitton, W.J., 2014. A geophysical perspective of value of information:
Examples of spatial decisions for groundwater sustainability. Environ. Syst. Decis.
34, 124–133. https://doi.org/10.1007/s10669-013-9487-9
Trainor-Guitton, W.J., Caers, J.K., Mukerji, T., 2011. A Methodology for Establishing a
Data Reliability Measure for Value of Spatial Information Problems. Math. Geosci.
43, 929–949. https://doi.org/10.1007/s11004-011-9367-0
Trainor-Guitton, W.J., Hoversten, G.M., Ramirez, A., Roberts, J., Juliusson, E., Key, K.,
Mellors, R., 2014. The value of spatial information for determining well placement:
A geothermal example. GEOPHYSICS 79, W27–W41.
References
57
https://doi.org/10.1190/geo2013-0337.1
Trainor-Guitton, W.J., Mukerji, T., Knight, R., 2013a. A methodology for quantifying
the value of spatial information for dynamic Earth problems. Stoch. Environ. Res.
Risk Assess. 27, 969–983. https://doi.org/10.1007/s00477-012-0619-4
Trainor-Guitton, W.J., Ramirez, A., Yang, X., Mansoor, K., Sun, Y., Carroll, S., 2013b.
Value of information methodology for assessing the ability of electrical resistivity
to detect CO2/brine leakage into a shallow aquifer. Int. J. Greenh. Gas Control 18,
101–113. https://doi.org/10.1016/j.ijggc.2013.06.018
Truex, M.J., 2018. Groundwater and Vadose Zone Remedy Management Strategies at
DOE Sites, in: SERDP ESTCP Symposium. Washington, D.C.
Truex, M.J., Johnson, T.C., Strickland, C.E., Peterson, J.E., Hubbard, S.S., 2013.
Monitoring Vadose Zone Desiccation with Geophysical Methods. Vadose Zo. J.
12. https://doi.org/10.2136/vzj2012.0147
Tsang, C.-F., Neretnieks, I., Tsang, Y., 2015. Hydrologic issues associated with nuclear
waste repositories. Water Resour. Res. 51, 1–50.
https://doi.org/10.1002/2014WR016259
Varouchakis, E.A., Palogos, I., Karatzas, G.P., 2016. Application of Bayesian and cost
benefit risk analysis in water resources management 534, 390–396.
Wainwright, H.M., Chen, J., Sassen, D.S., Hubbard, S.S., 2014. Bayesian hierarchical
approach and geophysical data sets for estimation of reactive facies over plume
scales. Water Resour. Res. 50, 4564–4584. https://doi.org/10.1002/2013WR013842
Wainwright, H.M., Flores Orozco, A., Bücker, M., Dafflon, B., Chen, J., Hubbard, S.S.,
Williams, K.H., 2016. Hierarchical Bayesian method for mapping biogeochemical
hot spots using induced polarization imaging. Water Resour. Res. 52, 533–551.
https://doi.org/10.1002/2015WR017763
Wallin, E.L., Johnson, T.C., Greenwood, W.J., Zachara, J.M., 2013. Imaging high stage
river-water intrusion into a contaminated aquifer along a major river corridor
using 2-D time-lapse surface electrical resistivity tomography. Water Resour. Res.
49, 1693–1708. https://doi.org/10.1002/wrcr.20119
Wang, J., Chen, L., Su, R., Zhao, X., 2018. The Beishan underground research laboratory
for geological disposal of high-level radioactive waste in China: Planning, site
selection, site characterization and in situ tests. J. Rock Mech. Geotech. Eng. 10,
411–435. https://doi.org/10.1016/j.jrmge.2018.03.002
Watson, D.B., Doll, W.E., Gamey, T.J., Sheehan, J.R., Jardine, P.M., 2005. Plume and
lithologic profiling with surface resistivity and seismic tomography. Ground
Water 43, 169–77. https://doi.org/10.1111/j.1745-6584.2005.0017.x
References
58
Webster, D.S., Proctor, J.F., Marine, I.W., 1970. Two -Well Tracer Test in Fractured
Crystalline Two-Well Tracer Test in Fractured Crystalline. Washington DC.
Williams, B.A., Thompson, M.D., Miller, S.F., 2012a. Integrated surface geophysical
investigation results at Liquid Efflucent Retention Facility, 200 East Area,
Washington. Richland, Washington.
Williams, B.A., Thompson, M.D., Miller, S.F., 2012b. Interpretation and integration of
seismic data in the Gable Gap. Richland, Washington.
Williams, B.A., Thompson, M.D., Miller, S.F., 2012c. Land streamer and gimbaled
geophones phase II-200 Areas: high-resolution seismic reflection suvey at the
Hanford Site. Richland, Washington.
Williams, B.A., Thompson, M.D., Miller, S.F., 2012d. Seismic reflection investigation at
the Liquid Effluent Retention Facility, 200 East Area, Hanford Site Richland,
Washington. Richland, Washington.
Williams, K.H., Kemna, A., Wilkins, M.J., Druhan, J., Arntzen, E., N’Guessan, A.L.,
Long, P.E., Hubbard, S.S., Banfield, J.F., 2009. Geophysical Monitoring of Coupled
Microbial and Geochemical Processes During Stimulated Subsurface
Bioremediation. Environ. Sci. Technol. 43, 6717–6723.
https://doi.org/10.1021/es900855j
Williams, M.D., Rockhold, M.L., Thorne, P.D., Chen, Y., 2008. Three-dimensional
groundwater models of the 300 Area at the Hanford Site, Washington State.
Richland, Washington.
Wu, W.M., Carley, J., Fienen, M., Mehlhorn, T., Lowe, K., Nyman, J., Luo, J., Gentile,
M.E., Rajan, R., Wagner, D., Hickey, R.F., Gu, B., Watson, D., Cirpka, O. a.,
Kitanidis, P.K., Jardine, P.M., Criddle, C.S., 2006. Pilot-scale in situ bioremediation
of uranium in a highly contaminated aquifer. 1. Conditioning of a treatment zone.
Environ. Sci. Technol. 40, 3978–3985. https://doi.org/10.1021/es051954y
Zhang, Y.-K., Neuman, S.P., 1990. A quasi-linear theory of non-Fickian and Fickian
subsurface dispersion: 2. Application to anisotropic media and the Borden site.
Water Resour. Res. 26, 903–913. https://doi.org/10.1029/WR026i005p00903
Zhao, Z., 2000. Application of geophysical exploration to site selection for waste
disposal of the Guangdong nuclear power plant. Geol. Prospect. (in Chinese) 36,
51–53.
Zheng, J., Zhang, Z., Dong, L., 2000. Environmental geophysics and its advance. Adv.
Earth Sci. (in Chinese) 15, 40–47.
Sources of uncertainties in electrical resistivity tomography (ERT): a review
59
3. Sources of uncertainties in electrical resistivity tomography
(ERT): a review
Manuscript prepared for journal submission as Tso, C.-H.M., Kuras, O., Binley, A.
(201x) Sources of uncertainties in electrical resistivity tomography (ERT): a review
Abstract
60
Abstract
The use of near-surface geophysics methods has become prevalent in many
areas, including environmental investigations, engineering, agriculture, archaeology,
forensic science, mineral exploration and hydrogeology. With improvements in
infrastructure, equipment, and modelling software and tools, near-surface geophysical
techniques have become more accessible. These developments allow the methods to
be readily used, in some cases, with limited background understanding of the
theoretical principles and limitations.
Like any other kind of measurement or interpretation, the near-surface
geophysical methods are subject to errors and uncertainties. To maximise the benefits
of such methods and reduce over-interpretation or failure to capture the maximum
information content they can provide, it is necessary to understand and manage the
various sources of errors and uncertainties, particularly to quantify them and assess
how they propagate through the various stages of interpretation. Using electrical
resistivity tomography (ERT) as an example, we review the various sources of
uncertainty and discuss approaches to handle them in practice. These sources include
measurement errors, inversion, petrophysical relationships, and application-specific
uncertainties.
In this review, we outline the advances made in characterizing and reducing
uncertainty along the various stages of the workflow and reveal that most approaches
focus their effort at a particular stage and rarely considers multiple sources of
uncertainties. Potential directions for future research includes: (i) reducing overall
prediction uncertainty through experimental design optimization; (ii) value of
information that considers uncertainty in the entire workflow; (iii) consider jointly
multiple sources of uncertainty and their effect downstream of the workflow; (iv)
propose alternative workflows to bypass certain high uncertainty steps in certain
applications.
Introduction
61
1) Introduction
Geophysical methods often provide a cost-effective and relatively rapid means
of collecting data on the properties of subsurface. The use of near-surface geophysics
methods has become prevalent and widely accessible due to improvements in
infrastructure, equipment, and modelling tools. These developments allow the
methods to be readily used, in some cases, with limited background understanding of
the theoretical principles and limitations. This draws many new users from different
fields. The geophysical properties returned from such surveys are often not the
quantity of interest (QoI) in these applications. Rather, the geophysical properties are
used as a proxy or are then converted to the QoI. For example, an electrical resistivity
tomography (ERT) survey returns a (smoothed or blocky) image of electrical resistivity
while monitoring the movement of a saline tracer (e.g. Cassiani et al., 2006; Kemna et
al., 2004; Wilkinson et al., 2010b). However, a hydrologist is interested in the soil
moisture content or solute concentration, which are converted from resistivity using
petrophysical relationships. Likewise, time-lapse resistivity images during CO2
injection monitoring needs to be converted to changes in CO2 concentrations (Carrigan
et al., 2013; Doetsch et al., 2013; Schmidt-Hattenberger et al., 2016; Yang et al., 2015).
Like any other kind of measurement or interpretation, the use of near-surface
geophysics involves errors and uncertainties. To maximise the benefits of using near-
surface geophysics there is a need, therefore, to understand and manage the various
sources of errors and uncertainties, in particular to quantify them and assess how they
propagate through the various stages of interpretation. The issue of uncertainty in
near-surface geophysics has become more relevant in recent years because these
methods have been increasingly used in a quantitative manner for prediction of
properties or state variables, or for decision making after being converted to QoIs. A
comprehensive appraisal of their various sources of uncertainties and their impact on
making predictions and decision making is desperately needed.
We focus our review on ERT, one of the most commonly used near surface
geophysical methods, with an emphasis on hydrogeophysics. In general, an ERT
survey follows the following five-step workflow: experimental design (i.e. desk study),
Measurement and data errors
62
data collection, inversion, interpretation, and prediction (Figure 1). While this serves
as a pipeline where information is passed along from measurements to prediction, it
also allows the propagation of uncertainties along it. Each part of the workflow
presents its own sources of uncertainties. Therefore, our review is organized as follows.
We will first discuss measurement and data errors in section 2 and inversion in section
3. We will then discuss interpretation of ERT results (section 4) before revisiting
inversion, this time focusing on incorporation of auxiliary information (section 5),
which is followed by a discussion on experimental design and emerging methods. We
then offer some conclusions and recommendations. Our review is limited to discussing
the various sources of uncertainties along the ERT workflow. Specialized reviews on
selected topics are available: for example, electrical imaging (Falzone et al., 2018), time-
lapse electrical imaging (Singha et al., 2015), geophysical imaging for surface water-
groundwater interaction and the critical zone (McLachlan et al., 2017; Parsekian et al.,
2015), landslide monitoring (Whiteley et al., 2019), and soil structure mapping
(Romero-Ruiz et al., 2018). A more general review on the uncertainty quantification in
hydrogeology and hydrogeophysics can be found in Linde et al. (2017).
Figure 1 Various sources of errors and uncertainties propagate through the ERT workflow (Binley et al.,
2015; Tran et al., 2016; Truex et al., 2013a). The workflow begins with experimental design, where the
objectives and details of the field campaign are laid out. It progresses to data collection in the field
using a data acquisition system. Then the data is inverted to obtain results in a usable format for
interpretations and discussions. Finally, the findings are used for decision making or prediction of
future events.
2) Measurement and data errors
In most ERT inversion approaches an objective function incorporating a weighted
misfit is used, allowing data value to be inversely weighted according to their errors.
It is important to recognise that such errors are not just measurement errors – they
Measurement and data errors
63
should also recognise modelling errors due to imperfections of the forward model to
represent the physics of electrical current flow.
2.1. Measurement errors and data quality
Measurement errors are often ignored (or poorly estimated) and yet they
propagate through the workflow. There are three ways to quantify measurement
errors: (i) stacking errors, which is returned by the ERT equipment based on averaging
the voltage signal caused by the injection of a low frequency alternating current; (ii)
repeatability errors, which are obtained by repeating the measurement sequence; (iii)
reciprocal errors, which are obtained by repeating measurement sequence but the
current and potential electrodes are swapped.
If we follow the classical approach and consider that all measurement errors
follow a normal distribution (i.e. assuming the measurements are independent and
identically distributed (i.i.d.) random variables and follow the central limit theorem),
then, for a given electrode quadrupole, each of the error quantities we obtain is only a
point on the normal distribution. Therefore, these errors needs to be fitted into an error
model to predict the error weights to be used in an inversion, and also to identify any
outlier data that one may wish to discard (or, ideally, repeat after addressing the source
of error, e.g. high contact resistance). Most error models recognize the proportionality
effect of ERT transfer resistance data, meaning the magnitude of the error in a transfer
resistance should be more or less proportional to the magnitude of the measurement.
It is important to note that such information on error trends cannot be established from
measurements expressed as apparent resistivity, since the measurements (and
associated errors) are scaled according to a geometrical factor. The simplest transfer
resistance error model, therefore, is a straight line between the transfer resistance and
the measurement errors (or variances) recorded (Binley et al., 1995; Slater et al., 2000).
This may be done in the linear, semi-log, or log-log space. The scatter in such a data-
error plot from such a process can make error model determination challenging. To
alleviate this issue, the data-error pairs may be grouped into logarithmically equally
distributed bins of transfer resistance before fitting (Koestel et al., 2008). It is normal to
consider measurement errors as uncorrelated. Recognizing the different electrodes
Measurement and data errors
64
used in a ERT survey may contribute differently to measurement errors; Tso et al. (2017)
devised a linear mixed effect model to fit a linear model to the errors, while using the
electrode number as an additional grouping variable.
There has been other studies of specific ERT applications focussed on
improving data quality. For example, Deceuster et al. (2013) developed an algorithm
to automatically detect changes in electrode contact properties for long-term
permanent ERT monitoring surveys. Mitchell and Oldenburg (2016) developed a new
data quality control methodology for large ERT datasets to identify and characterize
highly contaminated data from different noise sources. Lesparre et al. (2017) suggested
when performing difference inversion on time-lapse data (see section 3.4 for details),
reciprocal errors should be computed using difference in data between time steps.
Wilkinson et al. (2008) identified some electrode configurations that are highly
sensitive to geometric displacement in crosshole surveys.
One practical concern is that whether different ERT equipment gives
comparable response and whether their inverted resistivity results agree. The study of
Parsekian et al. (2017) compared six commercially available ERT equipment on the
same line of electrodes and showed that they give statistically similar apparent
resistivity results. They also showed by measuring the full waveform of a 4-electrode
array that systematic errors might be introduced due to poor electrode contact and
instrument-specific recording settings. In practice, some sources of systematic errors,
such as incorrectly swapping the connection of two electrodes, may be identified and
corrected in the data processing stage. In spite of the equipment used, at least one error
estimate should be recorded for error modelling and uncertainty propagation.
2.2. Forward modelling errors
In most ERT applications, current flows are not modelled analytically but are
approximated by finite difference or finite element models. Both methods require
discretization in space. The quality of the mesh (i.e. discretization) affects the quality
of the modelling of current flow and the discrepancy between the data generated by
the forward model and analytical solution (usually for a homogeneous domain) is a
forward modelling error. This can be especially important for surface surveys
Measurement and data errors
65
conducted in areas with rough or uneven topography (e.g. slopes). In general, the mesh
should be highly refined near electrode locations since the potential gradient is highest.
In field studies, the mesh should also extend laterally and downwards from the survey
area to satisfy the infinite earth assumption, or appropriate accounting should be made.
Forward modelling errors can also exist due to the effect of 3-D variability not
accounted for in a 2-D representation. Resistivity variation orthogonal to the survey
line (in a 2-D survey) can influence measurements but will not be correctly modelled
in a 2-D model. An example of this is where an ERT survey is used to monitor the
migration of a solute plume. Another example of 3-D effects is a borehole used for
deployment of electrodes in a cross-borehole configuration. Nimmer et al. (2008)
demonstrate the various scenarios where 3-D effects may be dominant in 2-D borehole
ERT studies (Figure 2): large diameter boreholes, borehole backfills with contrasting
conductivities to the formation, non-cantered targets, and heterogeneity outside the
imaging plane. Doetsch et al. (2010a) cautioned against the potential borehole-fluid
effects in borehole ERT studies, especially for shorter dipole spacing. Wagner et al.
(2015a) suggested using an explicit discretization of the borehole completion to
mitigate borehole-related effects for CO2 monitoring. Large forward modelling error is
most likely to occur in media containing complex fracture networks. To account for
this issue, a number of authors have proposed discrete fracture forward models as an
alternative (Beskardes and Weiss, 2018; Demirel et al., 2019; Roubinet and Irving, 2014)
to conventional finite element forward models. Electrode position errors are also
known to contribute to ERT data errors (Oldenborger et al., 2005), with recent work
focusing on accounting for movement of electrodes in landslide studies (Wilkinson et
al., 2016, 2010a). A further source of forward modelling error is failure to treat
electrodes as non-point sources or sensors, which can be significant for small scale
studies.
Inversion
66
Figure 2: Example of 3D borehole effects in 2D inversion reported in Nimmer et al. (2008), which shows
the 2.5D inversion results from Slater et al. (1997). The heavily fractured zone at around 25-m depth
can be seen as a low resistivity contrast to the background, but the large high resistivity (>2x105 Ω m)
in the centre of the image appears to be a result of the 2D resistivity model compensating for the low
resistivity along the borehole.
3) Inversion
In order to convert raw geophysical measurements (e.g. transfer resistances or
apparent resistivities) into useful products to interpret (e.g. spatial distribution of
electrical resistivity), data have to be converted to some parameter distribution that
can reproduce the measured data using a forward problem that describes the
underlying geophysical process, i.e. inversion. Inversion almost always suffer from ill-
posedness: Hadamard (1902) states that a well-posed problem must: (1) have a solution,
(2) the solution is unique, and (3) the solution’s behaviour changes continuously with
the initial conditions. Consequently, identifying a solution in the parameter space is
challenging.
When analysing a geophysical inverse problem, obtaining an optimal model is
usually not sufficient (Shi, 1998). Normally we also wish to have an estimation of
uncertainties and resolutions in the information content of the images. In other words,
we wish to know to what degree the inversion results represent the true (unknown)
structure. Any inversion procedure is considered to be incomplete without any
uncertainty or resolution analysis (ibid.). Taking a multi-hypothesis viewpoint, one
Inversion
67
would also hope to use geophysical data to falsify or corroborate hydrological models
(Linde, 2014). The Popper-Bayes philosophy proposed by Tarantola (2006) even argues
that data should be used first to falsify models. Regrettably, the above tasks have rarely
been undertaken due to heavy computational demand, difficulty in defining and
segregating sources of uncertainty, and the lack of framework to do so.
For most geophysical applications, the goal is to estimate spatial variability and
patterns of geophysical properties. This is challenging because of ill-posedness and
this is addressed either by regularization or computationally intensive Bayesian
inversion.
3.1. Regularized Inversion
The most common approach to tackling the problem of ill-posedness is to use
a Tikhonov-type regularization function, which makes such problems solvable by
minimizing the roughness of an image. Not surprisingly, the resultant images are
smooth. Philosophically, this approach can be considered as using an Occam’s razor
or adopting the law of parsimony, which argues “among competing hypotheses, the
one with the fewest assumptions should be selected” (note that “use the smoothest
image” is a strong assumption). This philosophy can be problematic in geophysical
inversion because maximum smoothness is often not justified, for example, in
fractured or layered systems, or when the subsurface is stimulated, e.g. by solute
injection.
To obtain the baseline resistivity structure, we seek to find a model solution
that minimizes the following objective function:
Φ = Φ𝑑 +Φ𝑚 = (𝑑 − 𝐹(𝑚))𝑇𝑊𝑑𝑇𝑊𝑑(𝑑 − 𝐹(𝑚)) + 𝛼𝑚
𝑇𝑅𝑚𝑚 (1)
where d are the data (e.g. measured apparent resistivities), F(m) is the set of simulated
data using the forward model and estimated parameters m. 𝑊𝑑 is a data weight matrix,
which, if we consider the uncorrelated measurement error case and ignore forward
model errors, is a diagonal matrix with entries equal to the reciprocal of the errors of
each measurement. Forward modelling errors are also added to the diagonal of 𝑊𝑑. 𝛼
Inversion
68
is the scalar regularisation factor, while 𝑅𝑚 is a roughness matrix that describes the
spatial connectedness of the parameter cell values.
Using a Gauss-Newton procedure (see derivation in Appendix 1), the above is
solved iteratively using the following solution:
(𝐽𝑇𝑊𝑑𝑇𝑊𝑑𝐽 + 𝛼𝑅𝑚)∆𝑚 = 𝐽𝑇𝑊𝑑
𝑇𝑊𝑑(𝑑 − 𝐹(𝑚)) − 𝛼𝑅𝑚𝑚𝑘 (2)
𝑚𝑘+1 = 𝑚𝑘 + ∆𝑚
where 𝐽 is the Jacobian (or sensitivity) matrix, given by 𝐽𝑖,𝑗 = 𝜕𝑑𝑖 𝜕𝑚𝑗⁄ ; 𝑚𝑘 is the
parameter set at iteration 𝑘 ; and ∆𝑚 is the parameter update at iteration 𝑘 . The
derivation of equation (2) can be found in the appendix. For the case of ERT, the inverse
problem is typically parameterized using log-transformed resistivities. The
computation of the Jacobian (or sensitivity) matrix 𝐽 can be a computationally
demanding task. Adjoint-state methods (Skyes et al., 1985) are normally applied to
greatly improve the computation of sensitivity matrices by efficiently computing the
gradient. Massive improvements have also been made in terms of computational
power. Highly parallelized codes (e.g. Johnson et al., 2010) have been developed to
greatly speed up computation by splitting matrix computations across multiple nodes.
3.2. Image appraisal
The model resolution matrix (R) is a matrix derived from the forward operator
of an inverse problem, which describes the quality of mapping in the model space. It
is given by:
𝐑 = (𝐽𝑇𝑊𝑑𝑇𝑊𝑑𝐽 + 𝛼𝑊𝑚
𝑇𝑊𝑚)−1𝐽𝑇𝑊𝑑
𝑇𝑊𝑑𝐽 (3)
A well-posed inverse problem would have a model resolution matrix that is a
diagonal matrix of 1. Strictly speaking, R can only be derived for linear inverse
problems. For weakly non-linear problems, it is common practice to approximate R by
linearization. Many methods have been proposed to evaluate the attributes of R for
image appraisal. Ramirez et al. (1995) used the diagonal elements of R as estimates of
spatial resolution of ERT images for cross-hole ERT surveys. More recently, Friedel
(2003) used the concept of averaging kernels (i.e. truncated singular value
Inversion
69
decomposition) to analyse attributes of R. Alumbaugh and Newman (2000) analysed
individual columns of R (also termed as PSFs for Point Spread Functions) in order to
estimate the spatial variation of the resolution in the horizontal and vertical directions
for 2-D and 3-D electromagnetic conductivity inversions. Their study was later
extended by Oldenborger and Routh (2009) to detect artefacts and study model
dependence of resolution. Day-Lewis and Lane (2004) and Day-Lewis et al. (2005) used
R combined with random field averaging and spatial statistics of the geophysical
property to predict the correlation loss between geophysical properties and
hydrological parameters. They showed that the correlation varies greatly within the
tomogram and direct application of petrophysical models to tomograms may yield
misleading estimates of hydrological properties. Caterina et al. (2013) reviewed the use
of R for image appraisal and edge detection techniques and proposed a new image
appraisal procedure that combines both methods. More details on edge detection is
given in section 4.1.
The computation of R is very expensive (Nolet et al., 1999; Oldenborger et al.,
2007) as it solves for the inverse of a matrix that contains the sums and products of
roughness and sensitivity matrices. R is a full matrix so sparse matrix solvers cannot
be used to improve computation efficiency. Alternative approaches are taken to
estimate resolution (or a proxy) at a considerably lower cost: some uses data error-
weighted cumulative sensitivity (Kemna, 2000), while other uses the depth-of-
investigation (DOI) concept (Oldenburg and Li, 1999). The latter essentially compares
how different the estimates are given two different prior models. These approaches
have inherent problems: for the prior, high sensitivity is not always correlated with
good resolution, while for the latter the DOI evaluates only a small subspace of R and
often sends false alarms of high or low resolutions.
Commonly used Tikhonov-type methods (i.e. regularizations), although
sufficient to produce a mathematically unique solution, do not really eliminate
uncertainty (Kitanidis, 2011). Their estimates are associated with errors, but these
errors are only measures of the discrepancy between the forward model and the “true”
Inversion
70
model, which does not contain much information on their uncertainty. Since the model
is deterministic, model structure uncertainty is assumed to be negligible.
In most practical cases, there exists information that may be helpful to better
solve an inverse problem in addition to the measurements. For example, the layering
of the subsurface from regional scale geological studies may be helpful to improve
estimates from a local ERT studies. Strictly speaking, such features cannot be
incorporated to the deterministic framework because unlike Bayesian methods, there
is no way within this framework to incorporate information that is not described by
the inverse model. The only option is to introduce some bias to the regularizations. For
example, one may force stronger regularization in one direction than the others, which
can be achieved at various degree of sophistication (Elwaseif and Slater, 2012;
Farquharson, 2008; Günther et al., 2006; Lelièvre and Farquharson, 2013). One may
even remove regularization in a certain region (e.g. Doetsch et al., 2012a; Johnson et al.,
2012b), such decision is usually informed from other data sources such as ground
penetrating radar (GPR) reflection data. This approach is common and practical as it
is a simple way to improve the estimates. In many cases, different regularization makes
dramatic changes in the estimate. An alternative to incorporate a priori geological
information is to use geostatistical regularization operators, which has recently been
extended to irregular mesh (Jordi et al., 2018). Sometimes regularization information
comes in the form of structural orientation information, which a recent study has
offered a couple of strategies to incorporate them in regularized inversion (Ross, 2004).
The problem with any regularization in deterministic inversion is that it is difficult to
evaluate whether its choice is justified. Even though the computation of sensitivity and
resolution matrix incorporates the regularization, these two measures only describe
the mapping of data on the model space. Again, a deterministic model assumes there
is no model uncertainty, which in many case this assumption is not justified.
A common misconception is to equate improved resolution to uncertainty
reduction—the two are not necessarily related. It is common in the literature to use
“uncertainty” and “resolution” interchangeably, as if they are the exact opposite.
Generally, resolution of an image is referred to its quality or its sharpness or its ability
Inversion
71
to identify features. Some geophysical literature, however, refers resolution analysis to
the analysis of non-uniqueness and uncertainty of solutions to inverse problems
(Gouveia and Scales, 1997; Mosegaard, 1999). Such a definition originates from the
Backus-Gilbert method (Backus and Gilbert, 1970, 1968)—the first attempt to quantify
uncertainty in geophysics because this method poses the inverse problem as finding
the optimal trade-off between variance (i.e. a measure for uncertainty) and spread (i.e.
a measure for spatial resolution). Subsequent articles are often not very clear whether
they are referring to the resolution of the recovered image (i.e. or model) or that of the
inverse problem. Even though subsequent work does not always link spatial resolution
and uncertainty, this confusion in terminology gives the wrong impression that the
two are necessarily the same. As an illustration of the problem, it is possible to get a
very sharp image of the subsurface properties but not know whether they are artefacts
or true features. Perhaps more critically, the sharper images may not change our ability
and confidence to answer the scientific questions of interest. On a related note, the
model resolution matrix (R) in geophysical inverse theory (Menke, 1989) is a measure
of mapping information from the data space to the model space. A more diagonal R
only means that such mapping is better. In the most extreme case, the subsurface can
be treated as homogeneous, making the inverse problem well-posed and thus well-
resolved but such formulation of the inverse problem will produce an image that does
not show any features of interest. Also, the computation of R requires the error level
of each data point to be known. Therefore, care must be taken to relate improved
resolution and uncertainty reduction. Nevertheless, the use of R is very useful to guide
and optimize geophysical survey design, which will be discussed in section 6.1.
Another metric to consider the uncertainty of inversion results is through the
examination of the posterior model covariance matrix MCM. There are two methods
to obtain the MCM. The first one is through computing it from the sensitivity matrix
obtained in the linearized inversion equations itself (Alumbaugh and Newman, 2000;
Ruggeri et al., 2014), which is given by
𝐌𝐂𝐌 = (𝐽𝑇𝑊𝑑𝑇𝑊𝑑𝐽 + 𝛼𝑊𝑚
𝑇𝑊𝑚)−1
(4)
Inversion
72
The second method is to perturb the data with its noise level and run Monte Carlo
simulations of the inversion (Aster et al., 2005; Tso et al., 2017). The MCM is given by
𝐌𝐂𝐌 =𝐀𝑇𝐀
𝑞 (5)
where 𝑞 is the number of realizations and rows of 𝐀 contains the difference between
the model estimate of each realizations and the average model. The diagonal terms are
the variances of the estimated model parameters, while the off-diagonal terms
represent smoothing between pairs of parameters. Examinination of the MCM can
provide insights to the uncertainty trade-off among different model parameters.
Notice that this method tends to return low uncertainty estimates in low resolution
areas (e.g. away from electrodes). The perturbed measurements mostly affect the high-
resolution region while the low-resolution regions are mainly controlled by
regularization, which causes the latter to show low variability. A variant of the Monte
Carlo approach is the data kit inversion (Fernández-Muñiz et al., 2019) which borrows
the idea of random sampling with replacement from bootstrapping. Instead of using
the entire dataset for inversion in each realization, a random subset of random size is
drawn at each realization. A similar idea of using bootstrapping for uncertainty
quantification was applied to the CO2 saturation samples derived directly from time-
lapse ERT at the Cranfield Pilot Site (Yang et al., 2014).
Uncertainty quantification of resistivity models recovered from deterministic
methods is one of the most poorly understood issues in geophysics. Only an inversion
based on a statistical approach provides a systematic framework to quantify such
uncertainties (Pankratov and Kuvshinov, 2015).
The regularized inversion suffers a number of shortcomings. Ideally, our goal
is not to obtain a unique solution (i.e. usually the smoothest model), but all the possible
models. In the next section, we change philosophy and introduce a number of other
approaches to invert ERT data.
Inversion
73
3.3. Global optimization approaches and Bayesian inversion
Regularized inversion smooths the model space to arrive at a single solution
which is a local minimum. This solution may not be a global minimum of the objective
function. In this case, global optimization approaches can be useful. This is usually
achieved by Monte Carlo methods such as particle swarm optimization or particle
filter (Fernández-Martínez et al., 2019). The advantage of these methods is that unlike
regularized inversion, it does not require the gradient of the problem, thus it does not
require the objective function to be differentiable nor the computation of the Jacobian
matrix.
Bayesian inversion relies on Bayes rule to relate prior model probabilities 𝑃(𝑚)
to posterior ones 𝑃(𝑚|) by a likelihood function 𝐿(𝑚|) = 𝑃(|𝑚) given observed
data :
𝑃(𝑚|) =𝑃(𝑚)𝑃(|𝑚)
𝑃() (6)
Assuming Gaussian probabilities, the likelihood function becomes
𝐿(𝑚|) ∝ exp (−1
2(𝑑 − 𝐆(𝑚))𝑇𝐶𝑑𝑑
−1(𝑑 − 𝐆(𝑚))) (7)
where 𝐆(𝑚) is the forward model, while the a priori model distribution becomes
𝑃(𝑚) ∝ exp (−1
2(𝑚 −𝑚𝑝𝑟𝑖𝑜𝑟)
𝑇𝐶𝑀−1(𝑚 −𝑚𝑝𝑟𝑖𝑜𝑟)) (8)
where 𝐶𝐷𝐷−1 and 𝐶𝑀
−1 are the data and model covariance matrices respectively. The
maximum a posteriori (MAP) solution 𝑚𝑀𝐴𝑃 is commonly computed, and it is where
the following objective function is minimized:
𝑚𝑀𝐴𝑃 = (𝑑 − 𝐆(𝑚))𝑇𝐶𝑑𝑑−1(𝑑 − 𝐆(𝑚)) + (𝑚 −𝑚𝑝𝑟𝑖𝑜𝑟)
𝑇𝐶𝑀−1(𝑚 −𝑚𝑝𝑟𝑖𝑜𝑟) (9)
The prior or model regularization is often simplified by replacing 𝐶𝑚 with a
matrix that quantifies the first derivative of the proposed model (flatness), the second
derivative (roughness) or its deviation from 𝑚𝑝𝑟𝑖𝑜𝑟 (damping) multiplied by a model
regularization weight. Notice that it is common practice for both the deterministic and
the Bayesian approach to apply model regularizations. However, they have important
distinctions. For the deterministic approach, regularization is applied directly to the
Inversion
74
estimated model. For the Bayesian approach, regularization is through the probability
density of the prior model assumed. Note that although there is striking resemblance
between the formulation of Tiknonov regularization and that of Bayesian MAP
inference and perhaps the solutions obtained are similar, the goals and interpretations
of the two methods are rather different (Shi, 1998). The former determine maximum
model smoothness that allows for data fitting, while the latter sample a probability
distribution of models that are consistent with both the data and prior information. In
fact the regularizing terms of the two (i.e. 𝑚𝑇𝑅𝑚𝑚 and 𝐶𝑀−1) can never be equal because
the inverse of 𝑚𝑇𝑅𝑚𝑚 is ill-posed.
Similar to the resolution matrix in deterministic methods, the resolution of the
inverse model is given by the posterior covariance matrix (Tarantola, 2005). However,
it is rarely computated because to define it often requires assumption such as Gaussian
prior probability density of the model, and linear relationship between model
parameters and data. Recently, Gunning et al. (2010) shows that resolution of
controlled-source electromagnetic (CSEM, a type of CSAMT method) data can be
inferred by either hierarchical models with free parameters for effective correlation
lengths (“Bayesian smoothing”), or model–choice frameworks applied to variable
resolution spatial models (“Bayesian splitting/merging”) (Kaipio and Somersalo, 2006).
Such evaluation of resolution for Bayesian inversion, however, has never been done
for near-surface geophysics applications.
Moving beyond the MAP solution, a major challenge for applying Bayesian
inversion rigorously is the evaluation of posterior distribution. In many cases, one
needs to approximate the full posterior probability distribution. Markov chain Monte
Carlo methods (McMC) build Markov chains in the parameter space formed of a
sequence of random variables that are drawn proportional to the posterior distribution
(i.e. importance sampling). Although many samples (or in many cases, model runs)
are needed, it nonetheless provides a means to sample high dimension and very
complicated posterior distributions accurately. More importantly, they allow
alternative model estimates to be identified, compared, ranked, and rejected. The
posterior distribution is obtained when a chain “forgets” its initial states and converge
Inversion
75
to a unique and stationary distribution. With some simplifications, the algorithm to
construct a McMC chain is outlined in Figure 3.
For geophysical applications, since complex model proposals are not described by a
functional form, the extended Metropolis rule (Mosegaard and Tarantola, 1995) is often
used to determine the acceptance probability 𝛂:
𝛂 = min 1,𝐿(𝑚𝑝𝑟𝑜𝑝,𝜂|)
𝐿(𝑚𝑐𝑢𝑟𝑟,𝜂| (10)
For more information on the Bayesian approach and McMC, the readers are referred
to Brunetti (2019).
1. Randomly select a point in the model space as the initial position, 𝑚0. Decide
on the total number of nstep steps to run.
FOR i = 1:nstep
2. Set 𝑚𝑐𝑢𝑟𝑟 = 𝑚𝑖−1
3. Propose a random move in the parameter space to generate 𝑚𝑝𝑟𝑜𝑝
4. Draw a value u from the uniform distribution between 0 and 1, 𝑈(0,1)
5. Accept or reject 𝑚𝑝𝑟𝑜𝑝based on an acceptance probability 𝜶: if 𝑈 < min(1, 𝛼)
then 𝑚𝑝𝑟𝑜𝑝 is accepted and the chain moves to the new position, 𝑚𝑖 = 𝑚𝑝𝑟𝑜𝑝;
otherwise, 𝑚𝑝𝑟𝑜𝑝 is rejected and the chain does not move, 𝑚𝑖 = 𝑚𝑐𝑢𝑟𝑟
ENDFOR
Figure 3 An outline of the Markov chain Monte Carlo (McMC) inversion algorithm.
An additional strength of Bayesian inversion is the ability to perform trans-
dimensional inversion. Conventional inversion algorithms require the user to define
the number of resistivity values to be estimated and the locations to which they
correspond. Trans-dimensional inversion, however, jointly estimates the number of
cells that discretize the model domain, sizes and shapes of the cells, and their resistivity
values. Generally, they utilize McMC algorithm for inversion and Metropolis-Hasting
Inversion
76
method for sampling the prior. Andersen et al. (2003) used a random coloured
polygonal model to jointly estimate shapes of different resistivity zones within a model
domain and the resistivity values of each zone. Sometimes it is preferable to consider
electrical resistivity changes, such as the delineation of plume caused by saline tracer
release. Ramirez et al (2005) used McMC to estimate the resistivity distribution from
the ratio of two ERT datasets. Using a base representation algorithm, their method
proposes subvolumes of varying sizes, shapes and resistivity values. Taking advantage
of Voronoi cells and the reversible-jump Markov chain Monte Carlo algorithm, Galetti
and Curtis (2018) developed a more robust and practical method for trans-dimensional
ERT inversion (TERT). They have also shown that TERT yield superior uncertainty
estimates than those obtained from conventional methods (e.g. resolution and
sensitivity matrix). Trans-dimensional inversion is advantageous over fixed-
dimensional inversion because it alleviates the “curse of dimensionality” (i.e.
computational demands of stochastically exploring higher‐dimensional spaces (i.e.
more unknowns) explode exponentially) (Curtis and Lomax, 2001; Scales and Snieder,
1997) and it takes advantage of the natural parsimony for Bayesian inference, meaning
the posterior models are only as complex as is required by data or prior information.
In a way, this is the result of a trade-off between Occam Factor (i.e. the ratio of posterior
to prior accessible volume) and high likelihood (low misfit). A mathematical proof of
this concept is given by Ray et al. (2018, 2016).
Bayesian methods treat uncertainties as probabilities and imply a
multiplicative relationship between different sources of uncertainty. Figure 4 shows
the framework of Bayesian inversion and their effect on uncertainty, as manifested in
the spread of probability density functions. In Tikhonov-type methods, all input
uncertainties are lumped to the starting model, making other sources of uncertainty
intractable. Bayesian methods, in contrast, allow one to disentangle or segregate
sources of uncertainty using multiplicative error models and sequential data
assimilation models (Rojas et al., 2008; Salamon and Feyen, 2010) or Bayesian
hierarchical models (Tsai and Elshall, 2013).
Inversion
77
Figure 4 Uncertainty quantification in Bayesian inversion (Iglesias and Stuart, 2014). The black dashed
lines and red solid lines denote prior and posterior probabilities. Essentially, the data drives an
updating of input probabilities (e.g. model parameters) and lead to an updating of model outputs,
including quantities of interest that are not directly observable from data.
3.4. Time-lapse inversion
The fast collection cycle of geophysical methods makes them suitable to
monitor subsurface processes. Sequences of images can be used to evaluate the
temporal evolution of subsurface properties. Hayley et al. (2011) summarizes the
various approaches used for time-lapse inversion. The most straightforward one is to
invert each image individually (i.e. absolute inversion). The initial model for starting
the inversion in absolute imaging is often the same for each time step, though the
conductivity model from the previous time step has also been used, as in Oldenborger
et al. (2007). The ratio inversion (Daily et al., 1992; Ramirez et al., 2005) inverts the ratio
of the resistivity between two times, while difference inversion (Labrecque and Yang,
2001) seeks to minimize the misfit between the difference in two datasets and the
difference between two model responses. The cascaded inversion (Miller et al., 2008;
Oldenborger et al., 2007) adds the constraint that the second inversion result should be
similar to the first. The 4-D inversion (Kim et al., 2009) seeks to invert data from all
times using a single system of equation. To do so, time regularization is added so that
changes from one time to another are smooth. A variety of the method is the active
time constraint approach (Karaoulis et al., 2011b, 2011a), which allows the time
regularization to vary depending on the degree of spatial resistivity changes occurring
between different monitoring stages. In simultaneous time-lapse inversion (Hayley et
al., 2011), time-lapse inversion for two time steps are done simultaneously and
constraints of smoothness and closeness to a reference model are applied to the
Inversion
78
difference image produced. It produces images of similar resolution to difference
inversion while needing less tailoring of regularization parameters. Following the
same idea of focusing (section 6.1), comparing R from different times also allow one to
anticipate areas where model parameters change most in time, thus allowing temporal
focusing (Wagner et al., 2015b; Wilkinson et al., 2015) . The feature is particularly
attractive for long-term monitoring systems where resistivity in only part of the
inversion domain is expected to change significantly over time. The above approaches
have all applied some smoothness constraints between datasets in the inversion and
they yield inferior results in zones where sensitivity is low and resistivity change is
abrupt. Hermans et al. (2016a) suggested a covariance-constrained difference
inversion, which computes the covariance between datasets and use it instead of
arbitrary smoothing functions to constrain the inversion.
An alternative approach to invert absolute time-lapse electrical resistivity data
is to use the extended Kalman filter (Lehikoinen et al., 2009; Mitchell et al., 2011; Nenna
et al., 2011). Kalman filter approaches, which have been successful for electrical
impedance tomography in medical imaging (Kaipio et al., 1999; Trigo et al., 2004;
Vauhkonen et al., 1998) and is rapidly gaining its popularity in hydrology (e.g. Reichle
et al., 2008; Schöniger et al., 2012; Xue and Zhang, 2014), allows inclusion of
information from previous time steps. Interesting properties of Bayesian filtering
approaches like Kalman filter include (i) both source terms and data can be taken as
uncertain; (ii) allows joint estimation of sources and model parameters. Ward et al.
(2014) combined Kalman filter and edge detection algorithms to track the movement
and evolution of saline tracer in 2-D and 3-D time-lapse experiments. Other methods
to improve time-lapse inversion includes time-series and time-frequency analysis
(Johnson et al., 2012a), as well as the data-domain correlation approach (Johnson et al.,
2009).
Resolution analysis and image appraisal methods for individual inversion are
also used for time-lapse inversion. Measures have not been developed specifically for
time-lapse inversion. For uncertainty analysis, one may follow either a frame-by-frame
analysis or a McMC approach as in individual inversion. The mean-log ratio (MLR) is
Interpretation, application, and prediction
79
commonly used to assess the amount of added information of a new image (Ramirez
et al., 2005; Trainor-Guitton et al., 2013b).
4) Interpretation, application, and prediction
4.1. Interpretation of ERT images
Interpretation of inverted ERT images is not straightforward because they are
smoothed representation of the actual system and their appearance are governed by
the inversion algorithm used. This tasks is further complicated when the resolution of
the survey is low. Dumont et al. (2018) studied the effects of smoothed resistivity
changes, pore water dilution, temperature, and initial water distribution on the
interpretation of infiltration depth, area, and volume at a landfill site and found that
the failure to consider these effects can lead to misestimating the infiltration metrics.
Carey et al. (2017) used forward modelling to investigate inversion artefacts resulting
from time-lapse ERT during rainfall simulations under three moisture contrast
scenarios and eight array configurations and showed that both factors contribute to
artefact development.
In some applications, the goal of conducting an ERT survey is to infer interfaces
from the resistivity images. Most studies identify the interfaces by visual interpretation,
while a few have attempted to do so using some automatic algorithms. The most
common one is the steepest gradient method (Chambers et al., 2013, 2012; Nguyen et
al., 2005), which assumes the interface is located at the point where resistivity is
steepest along a depth profile. Bouchedda et al. (2012) used parallel ERT and radar
travel time inversion and a Canny edge detector to exchange structural information
between them at every iteration to jointly estimate resistivity and slowness distribution.
Alternatively, an iso-resistivity surfaces approach can be used when the resistivity at
the interface is known through an independent measurement (Chambers et al., 2013).
Chambers et al. (2014b) applied a fuzzy c-means clustering approach (Ward et al., 2014)
to assign each cell within a ERT image to its most likely population and calculate the
resistivity at the interface. They found that it shows superior perfomance at a
Interpretation, application, and prediction
80
catchment site where the assumptions of the steepest gradient method break down.
Wainwright et al. (2016) took a multi-step approach to map the naturally-reduced zone
and the two hydrostratigraphic contacts at the Rifle IRFC site in Colorado. They first
inverted ERT and time-domain induced polarisation data collected from ten surface
lines. A Bayesian hierarchical analysis was then used to find correlations between the
inverted geophysical model and lithological logs information from 187 wells. There
exists another class of methods where the interface is inferred in the inversion and they
are discussed in section 5.1.
In some other applications, the goal of ERT is to calibrate hydrological models
and derive its parameters. Earlier studies focus on user ERT to capture tracer
breakthrough and derive vadose zone parameters (Binley et al., 2002a; Cassiani and
Binley, 2005). More recent study extends such use to more complex problems. For
example, Doetsch et al. (2013) found that including the reservoir boundaries as
structural constraints significantly improves the images of increasing supercritical CO2
saturation. They then used ERT-derived changes in subsurface electrical resistivity
along with gas composition data to constrain and calibrate hydrological models.
4.2. Petrophysical uncertainty
Petrophysical relationship converts geophysical data or parameter to
hydrological parameter of interest. They are largely considered as uncertain and can
impact any hydrological applications that uses geophysical data. A large number of
previous field and laboratory studies, however, has applied petrophysical
relationships deterministically (e.g. Chambers et al., 2014a; Dumont et al., 2018, 2016;
Miller et al., 2008; Wehrer and Slater, 2015). Among them, some studies compared
ERT-derived moisture contents with point-based moisture measurements (e.g. theta
probes, neutron probes, or Time domain reflectometry (TDR)) (Beff et al., 2013; Hübner
et al., 2015). Many review articles for hydrogeophysics and near-surface geophysics
have highlighted it as an outstanding challenge (Binley et al., 2015; Linde and Doetsch,
2016). Two studies have considered uncertain petrophysical relationships in their
analysis: Huisman et al. (2010) estimated a few Archie parameters alongside with other
parameter of interest, while Irving and Singha (2010) considered a range of Archie
Inversion continued
81
parameters in their McMC updating scheme. Recently, Tso et al. (2019) shows that the
natural variability in petrophysical parameters can lead to high uncertainty in
moisture content estimates derived from inverted ERT images.
There have been attempts to overcome this issue by estimating some form of
non-stationary petrophysical relations. Day-Lewis et al. (2005) uses a random field
averaging analysis, where an assumed covariance is used to describe the spatial
structure of the geophysical parameter and generate a realization for the geophysical
model, and together with an assumed petrophysical model, generate a realization of
the hydrological parameter of interest. Through analysing the model resolution matrix
of the geophysical inversion, random field averaging is used for upscaling and
generating ensemble variance tomograms and correlation coefficient between
estimated and true geophysical parameter. With them, probabilistic pixel-specific
petrophysical between hydrologic and estimated geophysical parameters can be
established for inversion of lab or field data. Moysey et al. (2005) and Singha and
Gorelick (2006) devises a full inverse statistical (FISt) calibration. Their method starts
with generating realizations of hydrogeological model. Passing the hydrogeological
models through a petrophysical model and geophysical forward model, realizations
of geophysical synthetic data are produced. Inverting each of the geophysical synthetic
data deterministically, the resultant images can be used to upscale hydrogeological
property models and develop non-stationary apparent petrophysical relations to better
estimate hydrologic properties.
5) Inversion continued
5.1. Incorporation of prior information and facies/interface detection
Introducing prior knowledge (or assumed knowledge) of the subsurface in the
inversion can have a dramatic effect on the resultant images: the spatial structure can
be significantly enhanced. This can be done in both the regularized inversion
framework where model roughness matrix is modified, and in the Bayesian
framework where prior knowledge of the spatial variability of properties can be
prescribed as prior probabilities (e.g. Andersen et al., 2003; Linde et al., 2006). The
Inversion continued
82
effects of incorporating prior information in inversion of ERT data is demonstrated by
Caterina et al. (2014), where they compare smoothness constrained inversion, reference
model inversion, structural inversion, and geostatistical inversion under various
settings.
Among Bayesian methods, geostatistical approaches (Linde et al., 2006; e.g. Yeh
et al., 2002) can be very robust and effective. They treat spatial properties as a random
field conditioned on data. For example, Linde et al. (2006) invert DC resistivity and
ground-penetrating radar traveltime data with a regularized least squares algorithm
but use stochastic regularization operators based on geostatistical models to constrain
the solution. The successive linear estimator (Yeh et al., 2002; Yeh and Liu, 2000) uses
a perturbation method to yield conditional expectation and the associate variance of
the subsurface conductivity field. At the end of each iteration, the covariance matrices
are updated. For the quasi-linear geostatistical approach (Kitanidis, 1995), the trend of
the data is updated instead. Kitanidis (2015a) reports that using the highly parallelized
forward code of Johnson et al. (2010) as a black box in lieu of the principal component
geostatistical approach (PCGA) (Lee and Kitanidis, 2014), which utilizes the leading
principal components from the prior covariance, a good approximation of the solution
can be obtained at a fraction of the cost. Latest approaches devise the use of multi-point
geostatistics (Hu and Chugunova, 2008; Linde et al., 2015a; Mariethoz and Caers, 2014)
or the Matérn covariance family in their inversion (Bouchedda et al., 2015). The
advantages of these methods is that they can better handle non-stationarity of spatial
variation and the geostatistics parameters can be estimated in the inversion. There is
potential to take advantage of the advances in ensemble Kalman inversion (Chada et
al., 2018), such as using a combination of level set methods, Matern covariance
functions and ensemble Kalman inversion for ERT data. The advantages of this
method are: (i.) can perform Bayesian inversion at a fraction of the cost of McMC, (ii.)
allows joint estimation of spatially varying correlation scales of the resistivity field, (iii.)
level set method allows the delineation of sharp contrast without the need to impose
constraints a priori.
Inversion continued
83
The roughness matrix in deterministic smoothness-constrained inversion can
be replaced by a model covariance matrix determined by geostatistical model
(Hermans et al., 2012). The approach yields superior results than conventional
inversion because it only smooths the model parameters to a level that is controlled by
the correlation length. The same approach can be extended to difference inversion of
time-lapse data (Hermans et al., 2016a). Alternatively, the model covariance matrix can
be determined using an image-guided approach. For example, structural information
from a seismic or GPR image can be extracted by first converting it to a greyscale
guiding image and then infer the coefficients of the model covariance matrix based on
structure oriented semblance (Zhou et al., 2014). Such approach has also been applied
to stochastic inversion (Zhou et al., 2016). It should also be noted that besides the
commonly used Tihonov regularization for smoothness-constrained inversion, the
total variation (TV) which favours a piecewise constant solution is another popular
choice. A total generalized variation scheme which alternate between the two is
recently proposed to avoid bias towards one of the two without any noticeable extra
computational cost (Sibbertt et al., 2017).
If there are known interfaces in the model domain, the roughness matrix can
be modified by disconnecting the regularization between two neighbouring model
parameters, with the option to prescribe known resistivity values. This approach is
particularly useful in fractured medium (Robinson et al., 2015) (see example in Figure
5), segregating known zones (e.g. bedrock and sediments) (Bazin and Pfaffhuber, 2013;
Coscia et al., 2011; Johnson et al., 2012b), and engineered barriers with known
boundaries (Slater and Binley, 2003). The regularization disconnect approach requires
prior knowledge of the interfaces. In some applications, there may be a sharp target in
the domain of which its location is not known (e.g. tracking a moving plume). In this
situation, focusing approaches that penalize changes in a certain way is preferred. One
example is the minimum gradient support (MGS) functional (Portniaguine and
Zhdanov, 1999), which minimizes the area where strong model variation and
discontinuity occur. An issue of this method is that the solution is very sensitive to the
MGS parameter. Nguyen et al. (2016) developed a data-driven approach to optimally
select the MGS parameter for time-lapse data. Finally, Fiandaca et al. (2015) proposed
Inversion continued
84
a generalized focusing method for time-lapse changes, which allows joint tuning of the
sharpness of the timeless changes and the size of the area/volume influenced by such
changes.
We have mentioned methods to interpret interfaces from ERT images (section
4.1). A potentially more reliable alternative is to estimate the interface within the ERT
data inversion. Irving and Singha (2010) used a sequential McMC approach for (i)
tracer concentration data, (ii) ERT data, and (iii) both to estimate the membership
probability of a binary facies systems. This approach can be extended to estimate
heterogeneous resistivity fields within each facies or zones using a Metropolis-within-
Gibbs method in which the model proposals are symmetric and the interface and
physical properties are updated alternately within the Gibbs framework (de Pasquale
et al., 2019; Iglesias et al., 2014). For a critical zone-bedrock problem (Figure 6), two
heterogeneous field for the entire model domain are generated (which are cropped to
match terrain) and then the two fields and the interface between them are sequentially
updated using a McMC algorithm. This resultant resistivity image clearly resolves
both the interface boundary and the in-zone heterogeneity.
Figure 5 This synthetic example shows that by disconnecting the smoothness constraint in a
regularized ERT inversion, the fracture network (red and yellow) is much better recovered (Robinson
et al., 2015). Compared with the smoothness constraint inversion, the smoothness disconnect case
shows pronounced elongated fractures and recover the very high conductivities along them.
Inversion continued
85
Figure 6 A new Bayesian inversion method that jointly estimates both the interface of the two units
and the sub-unit resistivity variations (de Pasquale et al., 2019). The approach estimates the resistivty
field of both unit (assuming they span the entire model domain) and their interface. The resultant field
is obtained by combining the two fields along the interface.
5.2. Coupled (joint and constrained) and uncoupled methods, and process-based
methods
The ultimate goal of geophysical inversion is rarely just the mapping of
geophysical properties. Usually it is used to gain understanding of other processes, for
instance, hydrogeology or hydrological processes. As shown already in section 4), the
process of interpreting an ERT image or to translate an ERT image to quantities of
interest can be a source of significant uncertainty. Coupled inversion or coupled
modelling can be an effective means to handle such issues because (i.) the inverse
problem is formulated in such a way that it returns the quantities of interests directly
and (ii.) usually the underlying process model that drives the geophysical response is
incorporated in the forward models used for inversion.
Traditionally, a geophysical image is used as a proxy for the spatial distribution
of interest. This approach is termed ‘un-coupled hydrogeophysical inversion’. An
alternative is to use a ‘coupled hydrogeophysical inversion’ (Hinnell et al., 2010;
Huisman et al., 2010; Irving and Singha, 2010; Kowalsky et al., 2005; Looms et al., 2008;
Inversion continued
86
Pollock and Cirpka, 2010), where feedback between geophysical and hydrology is
explicitly defined (as prior hydrological conceptualization). There are also situations
where multiple geophysical datasets are available to describe the same process. For
example, Lochbühler et al. (2013) used the structure-coupled approach (Doetsch, 2011;
Gallardo and Meju, 2011) to jointly invert multilevel crosshole slug interference tests,
temporal moments of tracer breakthrough curves, and cross-hole ground penetrating
radar data. Doetsch et al. (2010b) used the structural constraint to jointly invert ERT
and GPR data so that the geophysics-derived moisture content honours both
modalities. Huisman et al. (2010) used the DREAM algorithm (Vrugt et al., 2009),
which runs multiple Markov Chains simultaneously at the same time, to jointly use
ERT data and moisture content data from time-domain reflectometry (TDR) to estimate
soil hydraulic parameters of a model dykes. They also used AMALGAM (Vrugt and
Robinson, 2007) to approximate the Pareto front of parameter pairs. Jardani et al. (2013)
devised a stochastic McMC approach to invert hydraulic conductivity distribution
between two wells using ERT, self-potential and salt tracer concentration data jointly.
Notice that the application of coupled approaches is not limited to near-surface
processes. For example, transient pressure pulses and dc resistivity acquired at
permanent borehole sensors can be jointly inverted (Alpak et al., 2004). Similarly, the
combined use of electrical and gas composition data are used to constrain CO2
simulations (Doetsch et al., 2013).
When there are multiple geophysical data sets, the relationship between two
geophysical modalities may not be known for certain. In this case, the inversion may
benefit from not enforcing strict feedback between them. This is termed the
constrained inversion. Figure 7 shows the different strategies to invert multiple
geophysical data sets. Similarly, to relate geophysical and hydrological properties,
some petrophysical relationships are needed. These relationships can be determined
in the laboratory but they can be unreliable when applied to field problems. It may be
desirable to impose some “soft” process constraint instead so that uncertain inversion
results are related to one another in a more flexible sense. One idea is to use Monte
Carlo simulations using flow and transport models to generate training images that
are used as prior information to constrain geophysical inversion (Oware et al., 2013).
Inversion continued
87
While fully coupled approaches assume strict feedback between two modalities, and
individual inversion uses some possibly ambiguous regularization constraints,
process-based inversion can be viewed as an intermediate approach that allows
physically based regularization.
Figure 7 Illustration of different approaches for integrating multiple geophysical data sets for a
hydrogeological interpretation (Doetsch, 2011). Individual processing invert each dataset invidually
before interpreting them together qualitatively. Joint inversion invert all available datasets together—
explicit assumptions between different process models are required. A constrained inversion use parts
of the inversion results from one inversion to constrain another.
5.3. Data assimilation and data fusion of multiple data types
A straightforward way to enhance the information content in geophysical
measurements is to integrate hydrological and geophysical data from different sources.
This integration, however, is never as straightforward as it seems because different
data are measured in different spatial and temporal scales (Kitanidis, 2015b). This is a
problem because most integration would require the use of a Bayesian framework, but
the scales issues violate the “consistency” requirement for Bayesian updating.
Nevertheless, there have been many attempts to integrate data of different scales,
mostly relying on bringing all measurements to the same manageable scale. Ruggeri
et al. (2013) uses a two-step Bayesian sequential simulation approach: they first link
the low- and high-resolution geophysical data via a stochastic downscaling procedure,
followed by relating the downscaled geophysical data to the high-resolution hydraulic
conductivity distribution. JafarGandomi and Binley (2013) develop a trans-
dimensional geophysical data fusion workflow that first transform all geophysical data
into 1-D. Then they run stochastic joint inversion using Markov chain Monte Carlo
Discussion
88
methods to obtain discrete 1-D models. Subsequently, the information of the 1-D
models are quantified using entropy measures, which can be used to improve survey
design. Finally, the discrete 1-D models are fused using the Bayesian maximum
entropy approach (Christakos, 2000) to obtain 2-D or 3-D spatial distribution of
geophysical properties.
Localized averages can sometimes help assess and distinguish the information
embedded in different images. Miller and Routh (2007) and Routh (2009) use funnel
function analysis to provide a formalism to compute upper and lower bounds of
localized averages of the image. This framework can be helpful to determine scale
dependent (target-oriented) uncertainties taking into account the resolution of the
geophysical data. This idea can also be used probabilistically as adaptive kernel
smoothing to estimate the probability density of the data variable for each training
image at a much lower dimension (Hermans et al., 2015; Park et al., 2013).
As discussed in previous sections, prior information plays a dramatic role in
modifying inversion results. Better incorporation of prior information gives more and
better information for an inverse problem to start with. We should continue exploring
methods to improve incorporation of prior information. Moreover, the goal of
inversion is to obtain more accurate posterior estimates and their uncertainty bounds.
Better incorporation of prior and other sources of information means more information
to constrain the posterior solution, and thus it is key to the uncertainty quantification
in ERT.
6) Discussion
6.1. Experimental design
Once we understand the geophysical workflow and how errors and
uncertainties enter and propagate along it, this knowledge can be used to design
experiments with the aim to maximize the extraction of information or minimization
of uncertainty. Current work focusses mostly on inversion and examine whether
particular inversion algorithms and survey design to resolve the target feature.
Discussion
89
However, after understanding how uncertainty propagates along the ERT workflow,
future work should consider also other sources of uncertainty in experimental design.
For regularized inversion, the resolution matrix R can be used to confirm
whether a proposed survey can resolve model parameters of interest, as well as
ranking the merits of competing survey proposals (Friedel, 2003; Furman et al., 2007,
2004; Leube et al., 2012; Loke et al., 2014; Maurer et al., 2010). The ‘Compare R’ method
seeks to incrementally add measurements that would lead to the greatest resolution
improvement to the measurement scheme (Stummer et al., 2004; Wilkinson et al., 2012,
2006). Its variant combines the optimization for time-lapse studies (Wilkinson et al.,
2015) and electrode placement (Uhlemann et al., 2018). Similar approaches, some work
in the other direction (i.e. incrementally reduce measurements from a full set), can be
found in the literature (Blome et al., 2011; Maurer et al., 2000).
Bayesian approach also allows optimization of survey design. However, it does
not take a model resolution viewpoint as in deterministic inversion. Classical
approaches are based on the expected utility of data (de Barros and Rubin, 2008; Feyen
and Gorelick, 2005; Fu and Jaime Gómez-Hernández, 2009; Neuman et al., 2012; Shi et
al., 2013). Alternatives have been proposed such as the use of multi-objective
optimization (Nowak et al., 2015, 2012).
Existing literature only consider optimizing the design based on inversion
outputs. However, recognizing uncertainty stems from and propagates through the
ERT workflow (Figure 1), future work should strive to reduce the overall prediction
uncertainty instead and consider uncertainty sources at all steps along the workflow
(Figure 8). This approach can be used in conjunction with value of information (VOI)
analysis, which has been applied to ERT data previously (Nenna et al., 2011; Trainor-
Guitton et al., 2013b), to provide an estimate on whether the proposed field campaign
leads to added value or whether it is cost-effective.
Discussion
90
Figure 8 The ERT workflow as a pipeline for information and uncertainty propagation is helpful for
the experimental design of ERT surveys. It can be optimized to minimize uncertainty and maximize
the extraction of information.
6.2. Emerging methods and future work
The quantification of uncertainty in ERT is an important research topic
challenged by computational time constraints. Recently, Fernández-Martínez (2019)
developed an algorithm based on singular value decomposition and an exploratory
member of the particle swarm optimization family to speed up the estimation of ERT
model parameter uncertainty. Computation time reductions can also be achieved by
speeding up the forward model. Surrogate models or support vector machines can be
used to approximate model response at a fraction of the normal computation cost and
has been applied to many sub-disciplines of hydrology (Asher et al., 2015), yet there
has not been any application to near-surface geophysics.
The rapid advances of data science has provided new tools to approach the
inversion and uncertainty quantification problems (Scheidt et al., 2018). In particular,
the idea of Bayesian evidential learning is proposed as a new scientific protocol for
uncertainty quantification. It does not rely on traditional inversion methods but
instead rely on machine learning from Monte Carlo simulations. Meanwhile, some
authors argue for more geological realism in hydrogeophysics (Linde, 2014; Linde et
al., 2015b) and stress the importance of falsifying models that are inconsistent to data.
Machine learning are also applied to the estimation of moisture content variation from
ERT images (Moghadas and Badorreck, 2019). The authors built artificial neural
networks using the time series of soil moisture measurements at five depths and they
argue that the use of artificial neural networks better captures the nonlinear
Discussion
91
spatiotemporal behaviour of the wetting front than traditional petrophysical models
from the ERT data. With the rapid growth of automated ERT monitoring systems
deployed worldwide, key environmental data science challenges such as data
interoperability and real-time data analytics on cloud computing platforms will
become highly relevant.
This review has focused on using ERT as the main source of information.
However, it is noteworthy that ERT information are often used as “soft” data to
constrain other models, such as groundwater models. There is increased use of such
approach to constrain realizations of groundwater models generated using multi-point
geostatistics (Gottschalk et al., 2017; Hermans et al., 2015).
Recent work has extended coupled hydrogeophysical inversion to include even
more processes. In this situation, usually ERT data is used directly without inversion.
Tran et al. (2016) devised a coupled thermal-mechanical-hydrogeophysical inversion
to simultaneously estimate subsurface hydrological, thermal and petrophysical
parameters using hydrological, thermal and electrical resistivity tomography (ERT)
data. To reduce the number of unknown parameters, they screen the parameters using
global sensitivity analysis (Wainwright et al., 2014b) and fixed parameters that are
found to have negligible effect on the observation. The Morris (1991) method is
generally suitable for hydrogeophysical problems because of the relatively small
number of forward model runs required. More computationally intensive global
sensitivity analysis with the Sobol (2001) indices requires more forward runs and is
usually performed using surrogate modelling techniques (e.g. polynomial chaos
expansion), as demonstrated in a recent study on travel time of radionuclides (Deman
et al., 2016). There are a number studies that compares the two global sensitivity
analysis methods (e.g. Gan et al., 2014; Hermans et al., 2014). Tran et al. (2017) uses
soil liquid water content, temperature and electrical resistivity tomography (ERT) data
and a McMC inversion scheme to estimate the vertical distribution of soil organic
carbon content during a freeze-thaw event in the Arctic tundra. Even though the
models used to model ERT response are ever more complex, the principles discussed
above are relevant and can be used to guide future advances in data collection and
Conclusion
92
modelling. Another example is the coupled inversion of CO2 and hydraulic pressure,
CO2 arrival times and ERT data at the Ketzin site (Wagner and Wiese, 2018; Wiese et
al., 2018), where singular value decomposition and PEST (Christensen and Doherty,
2008) are used in combination to estimate parameters for a multi-physical reservoir
model.
7) Conclusion
We have reviewed the various sources of uncertainties in ERT along its
workflow. Each of the steps along the workflow presents its unique challenges to
represent and reduce its uncertainties, while these uncertainties propagate
downstream along the workflow. Significant effort has been put to characterize and
reduce uncertainty at each step. While majority of the literature has focused on
reducing inversion uncertainty, we also see an emergence of techniques focusing on
better handling of uncertainties in the interpretation or experimental design of ERT
surveys. We have shown that the inverse problem formulations for deterministic and
Bayesian inversion are similar but their meaning are not identical. We also highlight
that the ERT inverse problem, in general, is not a simple trade-off between resolution
and uncertainty. Future work needs to better address the apparently low uncertainty
(due to low resolution) away from electrodes in smoothness constrained inversions.
Incorporation of prior information, coupled inversions, and data assimilation are key
areas of research to improve the applicability and reliability of ERT results. Currently,
it is uncommon to consider uncertainty propagation along the entire ERT workflow.
Future work can focus on better describing such uncertainty propagation and perhaps
proposing alternative workflows to bypass certain high uncertainty steps.
Acknowledgement
This paper is published with the permission of the Executive Director of the British
Geological Survey (NERC) and Sellafield Ltd. (on behalf of the Nuclear
Decommissioning Authority). The first author is supported by a Lancaster University
Faculty of Science and Technology PhD studentship and a Nuclear Decommissioning
Authority PhD bursary.
References
93
References
Alpak, F., Torres-Verdin, C., Habashy, T.M., 2004. Joint inversion of transient pressure
and dc resistivity measurements acquired with in-situ permanent sensors: A
numerical study. Geophysics 69, 1173–1191. https://doi.org/10.1190/1.1801935
Alumbaugh, D.L., Newman, G.A., 2000. Image appraisal for 2-D and 3-D
electromagnetic inversion. Geophysics 65, 1455.
https://doi.org/10.1190/1.1444834
Andersen, K.E., Brooks, S.P., Hansen, M.B., 2003. Bayesian inversion of geoelectrical
resistivity data. J. R. Stat. Soc. Ser. B Stat. Methodol. 65, 619–642.
https://doi.org/10.1111/1467-9868.00406
Asher, M.J.J., Croke, B.F.W.F.W., Jakeman, A.J.J., Peeters, L.J.M.J.M., 2015. A review of
surrogate models and their application to groundwater modeling. Water Resour.
Res. 51, 5957–5973. https://doi.org/10.1002/2015WR016967
Aster, R., Borchers, B., Thurber, C.H., 2005. Parameter estimation and inverse problems.
Elsevier, Burlington, Massachusetts, USA.
Backus, G.E., Gilbert, F., 1970. Uniqueness in the Inversion of inaccurate Gross Earth
Data. Philos. Trans. R. Soc. London A 266, 123–192.
Backus, G.E., Gilbert, F., 1968. The Resolving power of Gross Earth Data. Geophys. J.
R. Astron. Soc. 16, 169–205.
Bazin, S., Pfaffhuber, A.A., 2013. Mapping of quick clay by electrical resistivity
tomography under structural constraint. J. Appl. Geophys. 98, 280–287.
https://doi.org/10.1016/J.JAPPGEO.2013.09.002
Beff, L., Günther, T., Vandoorne, B., Couvreur, V., Javaux, M., 2013. Three-dimensional
monitoring of soil water content in a maize field using Electrical Resistivity
Tomography. Hydrol. Earth Syst. Sci. 17, 595–609. https://doi.org/10.5194/hess-
17-595-2013
Beskardes, G.D., Weiss, C.J., 2018. Modelling dc responses of 3-D complex fracture
networks. Geophys. J. Int. 214, 1901–1912. https://doi.org/10.1093/gji/ggy234
Binley, A., Cassiani, G., Middleton, R., Winship, P., 2002. Vadose zone flow model
parameterisation using cross-borehole radar and resistivity imaging. J. Hydrol.
267, 147–159. https://doi.org/10.1016/S0022-1694(02)00146-4
Binley, A., Hubbard, S.S., Huisman, J.A., Revil, A., Robinson, D.A., Singha, K., Slater,
L.D., 2015. The emergence of hydrogeophsyics for improved understanding of
References
94
subsurface processes over multiple scales. Water Resour. Res. 51, 3837–3866.
https://doi.org/10.1002/2015WR017016
Binley, A., Ramirez, A.L., Daily, W., 1995. Regularised image reconstruction of noisy
electrical resistance data, in: Proceedings of the 4th Workshop of the European
Concerted Action on Process Tomography,. Bergen, pp. 401–410.
Blome, M., Maurer, H., Greenhalgh, S., 2011. Geoelectric experimental design —
Efficient acquisition and exploitation of complete pole-bipole data sets.
GEOPHYSICS 76, F15–F26. https://doi.org/10.1190/1.3511350
Bouchedda, A., Chouteau, M., Binley, A., Giroux, B., 2012. 2-D joint structural
inversion of cross-hole electrical resistance and ground penetrating radar data. J.
Appl. Geophys. 78, 52–67. https://doi.org/10.1016/j.jappgeo.2011.10.009
Bouchedda, A., Giroux, B., Gloaguen, E., 2015. Bayesian ERT Inversion using Non-
Stationary Inverse Matérn Covariance Matrix, in: SEG Annual Meeting. New
Orleans.
Brunetti, C., 2019. Bayesian model selection in hydrogeophysics and hydrogeology.
Université de Lausanne.
Carey, A.M., Paige, G.B., Carr, B.J., Dogan, M., 2017. Forward modeling to investigate
inversion artifacts resulting from time-lapse electrical resistivity tomography
during rainfall simulations. J. Appl. Geophys. 145, 39–49.
https://doi.org/10.1016/J.JAPPGEO.2017.08.002
Carrigan, C.R., Yang, X., LaBrecque, D.J., Larsen, D., Freeman, D., Ramirez, A.L., Daily,
W., Aines, R., Newmark, R., Friedmann, J., Hovorka, S., 2013. Electrical resistance
tomographic monitoring of CO2 movement in deep geologic reservoirs. Int. J.
Greenh. Gas Control 18, 401–408. https://doi.org/10.1016/j.ijggc.2013.04.016
Cassiani, G., Binley, A., 2005. Modeling unsaturated flow in a layered formation under
quasi-steady state conditions using geophysical data constraints. Adv. Water
Resour. 28, 467–477. https://doi.org/10.1016/j.advwatres.2004.12.007
Cassiani, G., Bruno, V., Villa, A., Fusi, N., Binley, A.M., 2006. A saline tracer test
monitored via time-lapse surface electrical resistivity tomography. J. Appl.
Geophys. 59, 244–259. https://doi.org/10.1016/j.jappgeo.2005.10.007
Caterina, D., Beaujean, J., Robert, T., Nguyen, F., 2013. A comparison study of different
image appraisal tools for electrical resistivity tomography. Near Surf. Geophys.
11, 639–657. https://doi.org/10.3997/1873-0604.2013022
References
95
Caterina, D., Hermans, T., Nguyen, F., 2014. Case studies of incorporation of prior
information in electrical resistivity tomography: comparison of different
approaches. Near Surf. Geophys. 12, 451–465. https://doi.org/10.3997/1873-
0604.2013070
Chada, N.K., Iglesias, M.A., Roininen, L., Stuart, A.M., 2018. Parameterizations for
ensemble Kalman inversion. https://doi.org/10.1088/1361-6420/aab6d9
Chambers, J.E., Gunn, D.A., Wilkinson, P.B., Meldrum, P.I., Haslam, E., Holyoake, S.,
Kirkham, M., Kuras, O., Merritt, A., Wragg, J., 2014a. 4D electrical resistivity
tomography monitoring of soil moisture dynamics in an operational railway
embankment. Near Surf. Geophys. 12, 61–72. https://doi.org/10.3997/1873-
0604.2013002
Chambers, J.E., Wilkinson, P.B., Penn, S., Meldrum, P.I., Kuras, O., Loke, M.H., Gunn,
D.A., 2013. River terrace sand and gravel deposit reserve estimation using three-
dimensional electrical resistivity tomography for bedrock surface detection. J.
Appl. Geophys. 93, 25–32. https://doi.org/10.1016/j.jappgeo.2013.03.002
Chambers, J.E., Wilkinson, P.B., Uhlemann, S., Sorensen, J.P.R., Roberts, C., Newell,
A.J., Ward, W.O.C., Binley, A., Williams, P.J., Gooddy, D.C., Old, G., Bai, L.,
2014b. Derivation of lowland riparian wetland deposit architecture using
geophysical image analysis and interface detection. Water Resour. Res. 50, 5886–
5905. https://doi.org/10.1002/2014WR015643
Chambers, J.E., Wilkinson, P.B., Wardrop, D., Hameed, A., Hill, I., Jeffrey, C., Loke,
M.H., Meldrum, P.I., Kuras, O., Cave, M., Gunn, D.A., 2012. Bedrock detection
beneath river terrace deposits using three-dimensional electrical resistivity
tomography. Geomorphology 177–178, 17–25.
https://doi.org/10.1016/j.geomorph.2012.03.034
Christakos, G., 2000. Modern Spatiotemporal Geostatistics. Oxford University Press,
New York.
Christensen, S., Doherty, J., 2008. Predictive error dependencies when using pilot
points and singular value decomposition in groundwater model calibration. Adv.
Water Resour. 31, 674–700. https://doi.org/10.1016/J.ADVWATRES.2008.01.003
Coscia, I., Greenhalgh, S.A., Linde, N., Doetsch, J., Marescot, L., Gunther, T., Vogt, T.,
Green, A.G., 2011. 3D crosshole ERT for aquifer characterization and monitoring
of infiltrating river water. Geophysics 76, G49–G59.
https://doi.org/10.1190/1.3553003
References
96
Curtis, A., Lomax, A., 2001. Prior information, sampling distributions, and the curse of
dimensionality. GEOPHYSICS 66, 372–378. https://doi.org/10.1190/1.1444928
Daily, W., Ramirez, A., LaBrecque, D., Nitao, J., 1992. Electrical resistivity tomography
of vadose water movement. Water Resour. Res. 28, 1429–1442.
https://doi.org/10.1029/91WR03087
Day-Lewis, F.D., Lane, J.W., 2004. Assessing the resolution-dependent utility of
tomograms for geostatistics. Geophys. Res. Lett. 31, 2–5.
https://doi.org/10.1029/2004GL019617
Day-Lewis, F.D., Singha, K., Binley, A.M., 2005. Applying petrophysical models to
radar travel time and electrical resistivity tomograms: Resolution-dependent
limitations. J. Geophys. Res. B Solid Earth 110, 1–17.
https://doi.org/10.1029/2004JB003569
de Barros, F.P.J., Rubin, Y., 2008. A risk-driven approach for subsurface site
characterization. Water Resour. Res. 44, 1–14.
https://doi.org/10.1029/2007WR006081
de Pasquale, G., Linde, N., Doetsch, J., Holbrook, W.S., 2019. Probabilistic inference of
subsurface heterogeneity and interface geometry using geophysical data.
Geophys. J. Int. https://doi.org/10.1093/gji/ggz055
Deceuster, J., Kaufmann, O., Camp, M. Van, 2013. Automated identification of changes
in electrode contact properties for long-term permanent ERT monitoring
experiments. Geophysics 78, E79–E94. https://doi.org/10.1190/GEO2012-0088.1
Deman, G., Konakli, K., Sudret, B., Kerrou, J., Perrochet, P., Benabderrahmane, H., 2016.
Using sparse polynomial chaos expansions for the global sensitivity analysis of
groundwater lifetime expectancy in a multi-layered hydrogeological model.
Reliab. Eng. Syst. Saf. 147, 156–169. https://doi.org/10.1016/j.ress.2015.11.005
Demirel, S., Irving, J., Roubinet, D., 2019. Comparison of REV size and tensor
characteristics for the electrical and hydraulic conductivities in fractured rock.
Geophys. J. Int. 216, 1953–1973. https://doi.org/10.1093/gji/ggy537
Doetsch, J., 2011. Joint and constrained inversion of geophsyical data for improved
imaging of aquier structure and processes. ETH Zurich.
Doetsch, J., Coscia, I., Greenhalgh, S., Linde, N., Green, A., Gunther, T., 2010a. The
borehole-fluid effect in electrical resistivity imaging. Geophysics 75, F107–F114.
https://doi.org/Doi 10.1190/1.3467824
References
97
Doetsch, J., Kowalsky, M.B., Doughty, C., Finsterle, S., Ajo-Franklin, J.B., Carrigan, C.R.,
Yang, X., Hovorka, S.D., Daley, T.M., 2013. Constraining CO2 simulations by
coupled modeling and inversion of electrical resistance and gas composition data.
Int. J. Greenh. Gas Control 18, 510–522. https://doi.org/10.1016/j.ijggc.2013.04.011
Doetsch, J., Linde, N., Binley, A., 2010b. Structural joint inversion of time-lapse
crosshole ERT and GPR traveltime data. Geophys. Res. Lett. 37, 1–6.
https://doi.org/10.1029/2010GL045482
Doetsch, J., Linde, N., Pessognelli, M., Green, A.G., Gunther, T., 2012. Constraining 3-
D electrical resistance tomography with GPR reflection data for improved
aquifer characterization. J. Appl. Geophys. 78, 68–76.
https://doi.org/10.1016/j.jappgeo.2011.04.008
Dumont, G., Pilawski, T., Dzaomuho-Lenieregue, P., Hiligsmann, S., Delvigne, F.,
Thonart, P., Robert, T., Nguyen, F., Hermans, T., 2016. Gravimetric water
distribution assessment from geoelectrical methods (ERT and EMI) in municipal
solid waste landfill. Waste Manag. 55, 129–140.
https://doi.org/10.1016/J.WASMAN.2016.02.013
Dumont, G., Pilawski, T., Hermans, T., Nguyen, F., Garré, S., 2018. The effect of initial
water distribution and spatial resolution on the interpretation of ERT monitoring
of water infiltration in a landfill cover. Hydrol. Earth Syst. Sci. Discuss. 1–26.
https://doi.org/10.5194/hess-2018-163
Elwaseif, M., Slater, L., 2012. Improved Resistivity Imaging of Targets with Sharp
Boundaries Using an Iterative Disconnect Procedure. J. Environ. Eng. Geophys.
17, 89–101. https://doi.org/10.2113/JEEG17.2.89
Falzone, S., Robinson, J., Slater, L., 2018. Characterization and Monitoring of Porous
Media with Electrical Imaging: A Review. Transp. Porous Media 1–26.
https://doi.org/10.1007/s11242-018-1203-2
Farquharson, C.G., 2008. Constructing piecewise-constant models in multidimensional
minimum-structure inversions. GEOPHYSICS 73, K1–K9.
https://doi.org/10.1190/1.2816650
Fernández-Martínez, J.L., Fernández-Muñiz, Z., Xu, S., Cernea, A., Sirieix, C., Riss, J.,
2019. Efficient uncertainty analysis of the 3D electrical tomography inverse
problem. GEOPHYSICS 1–62. https://doi.org/10.1190/geo2017-0729.1
Fernández-Muñiz, Z., Khaniani, H., Fernández-Martínez, J.L., 2019. Data kit inversion
and uncertainty analysis. J. Appl. Geophys. 161, 228–238.
https://doi.org/10.1016/j.jappgeo.2018.12.022
References
98
Feyen, L., Gorelick, S.M., 2005. Framework to evaluate the worth of hydraulic
conductivity data for optimal groundwater resources management in
ecologically sensitive areas. Water Resour. Res. 41, 1–13.
https://doi.org/10.1029/2003WR002901
Fiandaca, G., Doetsch, J., Vignoli, G., Auken, E., 2015. Generalized focusing of time-
lapse changes with applications to direct current and time-domain induced
polarization inversions 1101–1112. https://doi.org/10.1093/gji/ggv350
Friedel, S., 2003. Resolution, stability and efficiency of resistivity tomography
estimated from a generalized inverse approach. Geophys. J. Int. 153, 305–316.
https://doi.org/10.1046/j.1365-246X.2003.01890.x
Fu, J., Jaime Gómez-Hernández, J., 2009. Uncertainty assessment and data worth in
groundwater flow and mass transport modeling using a blocking Markov chain
Monte Carlo method. J. Hydrol. 364, 328–341.
https://doi.org/10.1016/j.jhydrol.2008.11.014
Furman, A., Ferre, T.P. a., Heath, G.L., 2007. Spatial focusing of electrical resistivity
surveys considering geologic and hydrologic layering. Geophysics 72, F65.
https://doi.org/10.1190/1.2433737
Furman, A., Ferré, T.P.A., Warrick, A.W., 2004. Optimization of ERT Surveys for
Monitoring Transient Hydrological Events Using Perturbation Sensitivity and
Genetic Algorithms. Vadose Zo. J. 3, 1230. https://doi.org/10.2136/vzj2004.1230
Galetti, E., Curtis, A., 2018. Transdimensional Electrical Resistivity Tomography. J.
Geophys. Res. Solid Earth. https://doi.org/10.1029/2017JB015418
Gallardo, L.A., Meju, M.A., 2011. Structure-coupled multiphysics imaging in
geophysical sciences. Rev. Geophys. 49, RG1003.
https://doi.org/10.1029/2010RG000330.24
Gan, Y., Duan, Q., Gong, W., Tong, C., Sun, Y., Chu, W., Ye, A., Miao, C., Di, Z., 2014.
A comprehensive evaluation of various sensitivity analysis methods: A case
study with a hydrological model. Environ. Model. Softw. 51, 269–285.
https://doi.org/10.1016/j.envsoft.2013.09.031
Gottschalk, I.P., Hermans, T., Knight, R., Caers, J., Cameron, D.A., Regnery, J., McCray,
J.E., 2017. Integrating non-colocated well and geophysical data to capture
subsurface heterogeneity at an aquifer recharge and recovery site. J. Hydrol. 555,
407–419. https://doi.org/10.1016/j.jhydrol.2017.10.028
References
99
Gouveia, W.P., Scales, J. a, 1999. Resolution of seismic waveform inversion: Bayes
versus Occam. Inverse Probl. 13, 323–349. https://doi.org/10.1088/0266-
5611/13/2/009
Gunning, J., Glinsky, M.E., Hedditch, J., 2010. Resolution and uncertainty in 1D CSEM
inversion: A Bayesian approach and open-source implementation. Geophysics
75, F151. https://doi.org/10.1190/1.3496902
Günther, T., Rücker, C., Spitzer, K., 2006. Three-dimensional modelling and inversion
of dc resistivity data incorporating topography - II. Inversion. Geophys. J. Int.
166, 506–517. https://doi.org/10.1111/j.1365-246X.2006.03011.x
Hadamard, J., 1902. Sur les problèmes aux dérivées partielles et leur signification
physique. Princet. Univ. Bull. 13, 49–52.
Hayley, K., Pidlisecky, A., Bentley, L.R., 2011. Simultaneous time-lapse electrical
resistivity inversion. J. Appl. Geophys. 75, 401–411.
https://doi.org/10.1016/j.jappgeo.2011.06.035
Hermans, T., Kemna, A., Nguyen, F., 2016. Covariance-constrained difference
inversion of time-lapse electrical resistivity tomography data. GEOPHYSICS 81,
E311–E322. https://doi.org/10.1190/geo2015-0491.1
Hermans, T., Nguyen, F., Caers, J., 2015. Uncertainty in training image-based inversion
of hydraulic head data constrained to ERT data:Workflow and case study. Water
Resour. Res. 51, 5332–5352. https://doi.org/10.1002/2014WR016259
Hermans, T., Nguyen, F., Robert, T., Revil, A., 2014. Geophysical methods for
monitoring temperature changes in shallow low enthalpy geothermal systems.
Energies 7, 5083–5118. https://doi.org/10.3390/en7085083
Hermans, T., Vandenbohede, A., Lebbe, L., Martin, R., Kemna, A., Beaujean, J., Nguyen,
F., 2012. Imaging artificial salt water infiltration using electrical resistivity
tomography constrained by geostatistical data. J. Hydrol. 438–439, 168–180.
https://doi.org/10.1016/J.JHYDROL.2012.03.021
Hinnell, A.C., Ferr, T.P.A., Vrugt, J.A., Huisman, J.A., Moysey, S., Rings, J., Kowalsky,
M.B., 2010. Improved extraction of hydrologic information from geophysical
data through coupled hydrogeophysical inversion. Water Resour. Res. 46, 1–14.
https://doi.org/10.1029/2008WR007060
Hu, L.Y., Chugunova, T., 2008. Multiple-point geostatistics for modeling subsurface
heterogeneity: A comprehensive review. Water Resour. Res. 44, 1–14.
https://doi.org/10.1029/2008WR006993
References
100
Hübner, R., Heller, K., Günther, T., Kleber, A., 2015. Monitoring hillslope moisture
dynamics with surface ERT for enhancing spatial significance of hydrometric
point measurements. Hydrol. Earth Syst. Sci. 19, 225–240.
https://doi.org/10.5194/hess-19-225-2015
Huisman, J.A., Rings, J., Vrugt, J.A., Sorg, J., Vereecken, H., 2010. Hydraulic properties
of a model dike from coupled Bayesian and multi-criteria hydrogeophysical
inversion. J. Hydrol. 380, 62–73. https://doi.org/10.1016/j.jhydrol.2009.10.023
Iglesias, M.A., Lin, K., Stuart, A.M., 2014. Well-posed Bayesian geometric inverse
problems arising in subsurface flow. Inverse Probl. 30, 114001.
https://doi.org/10.1088/0266-5611/30/11/114001
Iglesias, M.A., Stuart, A.M., 2014. Inverse problems and uncertainty quantification.
SIAM News 2–3.
Irving, J., Singha, K., 2010. Stochastic inversion of tracer test and electrical geophysical
data to estimate hydraulic conductivities. Water Resour. Res. 46, W11514.
https://doi.org/10.1029/2009WR008340
JafarGandomi, A., Binley, A., 2013. A Bayesian trans-dimensional approach for the
fusion of multiple geophysical datasets. J. Appl. Geophys. 96, 38–54.
https://doi.org/10.1016/j.jappgeo.2013.06.004
Jardani, A., Revil, A., Dupont, J.P., 2013. Stochastic joint inversion of hydrogeophysical
data for salt tracer test monitoring and hydraulic conductivity imaging. Adv.
Water Resour. 52, 62–77. https://doi.org/10.1016/j.advwatres.2012.08.005
Johnson, T.C., Slater, L.D., Ntarlagiannis, D., Day-Lewis, F.D., Elwaseif, M., 2012a.
Monitoring groundwater-surface water interaction using time-series and time-
frequency analysis of transient three-dimensional electrical resistivity changes.
Water Resour. Res. 48, 1–13. https://doi.org/10.1029/2012WR011893
Johnson, T.C., Versteeg, R.J., Huang, H., Routh, P.S., 2009. Data-domain correlation
approach for joint hydrogeologic inversion of time-lapse hydrogeologic and
geophysical data. Geophysics 74, F127. https://doi.org/10.1190/1.3237087
Johnson, T.C., Versteeg, R.J., Rockhold, M., Slater, L.D., Ntarlagiannis, D., Greenwood,
W.J., Zachara, J., 2012b. Characterization of a contaminated wellfield using 3D
electrical resistivity tomography implemented with geostatistical, discontinuous
boundary, and known conductivity constraints. Geophysics 77, EN85.
https://doi.org/10.1190/geo2012-0121.1
Johnson, T.C., Versteeg, R.J., Ward, A., Day-Lewis, F.D., Revil, A., 2010. Improved
hydrogeophysical characterization and monitoring through parallel modeling
References
101
and inversion of time-domain resistivity and induced-polarization data.
Geophysics 75, WA27. https://doi.org/10.1190/1.3475513
Jordi, C., Doetsch, J., Günther, T., Schmelzbach, C., Robertsson, J.O., 2018.
Geostatistical regularization operators for geophysical inverse problems on
irregular meshes. Geophys. J. Int. 213, 1374–1386.
https://doi.org/10.1093/gji/ggy055
Kaipio, J.P., Karjalainen, P.A., Somersalo, E., Vauhkonen, M., 1999. State estimation in
time-varying electrical impedance tomography. Ann. N. Y. Acad. Sci.
https://doi.org/10.1111/j.1749-6632.1999.tb09492.x
Kaipio, J.P., Somersalo, E., 2006. Statistical and computational inverse problems.
Springer-Verlag, New York.
Karaoulis, M., Kim, J.-H., Tsourlos, P.I., 2011a. 4D active time constrained resistivity
inversion. J. Appl. Geophys. 73, 25–34.
https://doi.org/10.1016/j.jappgeo.2010.11.002
Karaoulis, M., Revil, A., Werkema, D.D., Minsley, B.J., Woodruff, W.F., Kemna, A.,
2011b. Time-lapse three-dimensional inversion of complex conductivity data
using an active time constrained (ATC) approach. Geophys. J. Int. 187, 237–251.
https://doi.org/10.1111/j.1365-246X.2011.05156.x
Kemna, A., 2000. Tomographic inversion of complex resistivity. Ruhr-Universität
Bochum.
Kemna, A., Binley, A.M., Slater, L., 2004. Cross-borehole IP imaging for engineering
and environmental applications. Geophysics 69, 97–107.
https://doi.org/10.1190/1.1649379
Kim, J.-H., Yi, M.-J., Park, S.-G., Kim, J.G., 2009. 4-D inversion of DC resistivity
monitoring data acquired over a dynamically changing earth model. J. Appl.
Geophys. 68, 522–532. https://doi.org/10.1016/j.jappgeo.2009.03.002
Kitanidis, P.K., 2015a. Three-dimensional ERT imaging by the geostatistical approach,
in: American Geophysical Union Fall Meeting. San Francisco, California.
Kitanidis, P.K., 2015b. Persistent questions of heterogeneity, uncertainty, and scale in
subsurface flow and transport. Water Resour. Res. 51, 5888–5904.
https://doi.org/10.1002/2015WR017639
Kitanidis, P.K., 2011. Bayesian and Geostatistical Approaches to Inverse Problems, in:
Biegler, L., Biros, G., Ghattas, O., Heinkenschloss, M., Keyes, D., Mallick, B.,
Marzouk, Y., Tenorio, L., van Bloemen Waanders, B., Willcox, K. (Eds.), Large-
References
102
Scale Inverse Problems and Quantification of Uncertainty. John Wiley & Sons,
Ltd, Chichester, UK, pp. 71–85. https://doi.org/10.1002/9780470685853
Kitanidis, P.K., 1995. Quasi-linear geostatistical theory for inversing. Water Resour.
Res. 31, 2411–2419.
Koestel, J., Kemna, A., Javaux, M., Binley, A., Vereecken, H., 2008. Quantitative
imaging of solute transport in an unsaturated and undisturbed soil monolith
with 3-D ERT and TDR. Water Resour. Res. 44, 1–17.
https://doi.org/10.1029/2007WR006755
Kowalsky, M.B., Finsterle, S., Peterson, J., Hubbard, S., Rubin, Y., Majer, E., Ward, A.,
Gee, G., 2005. Estimation of field-scale soil hydraulic and dielectric parameters
through joint inversion of GPR and hydrological data. Water Resour. Res. 41, n/a-
n/a. https://doi.org/10.1029/2005WR004237
Labrecque, D.J., Yang, X., 2001. Difference inversion of ERT data: a fast inversion
method for 3-D in situ monitoring. J. Environ. Eng. Geophys. 6, 83–89.
https://doi.org/dx.doi.org/10.4133/JEEG6.2.83
Lee, J., Kitanidis, P.K., 2014. Large-scale hydraulic tomography and joint inversion of
head and tracer data using the Principal Component Geostatistical Approach
(PCGA). Water Resour. Res. 50. https://doi.org/10.1002/2014WR015483
Lehikoinen, A., Finsterle, S., Voutilainen, A., Kowalsky, M.B., Kaipio, J.P., 2009.
Dynamical inversion of geophysical ERT data: state estimation in the vadose
zone. Inverse Probl. Sci. Eng. 17, 715–736.
https://doi.org/10.1080/17415970802475951
Lelièvre, P.G., Farquharson, C.G., 2013. Gradient and smoothness regularization
operators for geophysical inversion on unstructured meshes. Geophys. J. Int. 195,
330–341. https://doi.org/10.1093/gji/ggt255
Lesparre, N., Nguyen, F., Kemna, A., Robert, T., Hermans, T., Daoudi, M., Flores-
Orozco, A., 2017. A new approach for time-lapse data weighting in electrical
resistivity tomography. GEOPHYSICS 82, E325–E333.
https://doi.org/10.1190/geo2017-0024.1
Leube, P.C., Geiges, A., Nowak, W., 2012. Bayesian assessment of the expected data
impact on prediction confidence in optimal sampling design. Water Resour. Res.
48, 1–16. https://doi.org/10.1029/2010WR010137
Linde, N., 2014. Falsification and corroboration of conceptual hydrological models
using geophysical data. Wiley Interdiscip. Rev. Water 1, 151–171.
https://doi.org/10.1002/wat2.1011
References
103
Linde, N., Binley, A., Tryggvason, A., Pedersen, L.B., Revil, A., 2006. Improved
hydrogeophysical characterization using joint inversion of cross-hole electrical
resistance and ground-penetrating radar traveltime data. Water Resour. Res. 42,
WR005131. https://doi.org/10.1029/2006WR005131
Linde, N., Doetsch, J., 2016. Joint Inversion in Hydrogeophysics and Near-Surface
Geophysics, in: Moorkamp, M., Lelievre, P., Linde, N., Khan, A. (Eds.),
Integrated Imaging of the Earth. John Wiley & Sons, Inc., Hoboken, New Jersey,
pp. 117–135. https://doi.org/10.1002/9781118929063.ch7
Linde, N., Ginsbourger, D., Irving, J., Nobile, F., Doucet, A., 2017. On uncertainty
quantification in hydrogeology and hydrogeophysics. Adv. Water Resour. 110,
166–181. https://doi.org/10.1016/J.ADVWATRES.2017.10.014
Linde, N., Lochbühler, T., Dogan, M., Van Dam, R.L., 2015a. Tomogram-based
comparison of geostatistical models: Application to the Macrodispersion
Experiment (MADE) site. J. Hydrol. 531, 543–556.
https://doi.org/10.1016/j.jhydrol.2015.10.073
Linde, N., Renard, P., Mukerji, T., Caers, J., 2015b. Geological Realism in
Hydrogeological and Geophysical Inverse Modeling: a Review. Adv. Water
Resour. 86, 86–101. https://doi.org/10.1016/j.advwatres.2015.09.019
Lochbühler, T., Doetsch, J., Brauchler, R., Linde, N., 2013. Structure-coupled joint
inversion of geophysical and hydrological data. Geophysics 78, ID1–ID14.
https://doi.org/10.1190/geo2012-0460.1
Loke, M.H., Wilkinson, P.B., Uhlemann, S.S., Chambers, J.E., Oxby, L.S., 2014.
Computation of optimized arrays for 3-D electrical imaging surveys. Geophys. J.
Int. 199, 1751–1764. https://doi.org/10.1093/gji/ggu357
Looms, M.C., Binley, A., Jensen, K.H., Nielsen, L., Hansen, T.M., 2008. Identifying
Unsaturated Hydraulic Parameters Using an Integrated Data Fusion Approach
on Cross-Borehole Geophysical Data. Vadose Zo. J. 7, 238.
https://doi.org/10.2136/vzj2007.0087
Mariethoz, G., Caers, J., 2014. Multiple-Point Geostatistics. John Wiley & Sons, Ltd,
Chichester, UK.
Maurer, H., Boerner, D.E., Curtis, A., 2000. Design strategies for electromagnetic
geophysical surveys. Inverse Probl. 16, 1097–1117. https://doi.org/10.1088/0266-
5611/16/5/302
Maurer, H., Curtis, A., Boerner, D.E., 2010. Recent advances in optimized geophysical
survey design. Geophysics 75, A177--A194. https://doi.org/10.1190/1.3484194
References
104
McLachlan, P.J., Chambers, J.E., Uhlemann, S.S., Binley, A., 2017. Geophysical
characterisation of the groundwater–surface water interface. Adv. Water Resour.
109, 302–319. https://doi.org/10.1016/j.advwatres.2017.09.016
Menke, W., 1989. Geophysical Data Analysis: Discrete Inverse Theory. Academic Press.
Miller, C.R., Routh, P.S., 2007. Resolution analysis of geophysical images: Comparison
between point spread function and region of data influence measures. Geophys.
Prospect. 55, 835–852. https://doi.org/10.1111/j.1365-2478.2007.00640.x
Miller, C.R., Routh, P.S., Brosten, T.R., McNamara, J.P., 2008. Application of time-lapse
ERT imaging to watershed characterization. Geophysics 73, G7.
https://doi.org/10.1190/1.2907156
Mitchell, M.A., Oldenburg, D.W., 2016. Data quality control methodology for large ,
non-conventional DC resistivity datasets. J. Appl. Geophys. 3, 948–953.
https://doi.org/10.1016/j.jappgeo.2016.09.018
Mitchell, V., Knight, R., Pidlisecky, A., 2011. Inversion of time-lapse electrical
resistivity imaging data for monitoring infiltration. Lead. Edge 140–144.
Moghadas, D., Badorreck, A., 2019. Machine learning to estimate soil moisture from
geophysical measurements of electrical conductivity. Near Surf. Geophys.
https://doi.org/10.1002/nsg.12036
Morris, M.D., 1991. Factorial Sampling Plans for Preliminary Computational
Experiments, Technometrics.
Mosegaard, K., 1999. Resolution analysis of general inverse problems through inverse
Monte Carlo sampling. Inverse Probl. 14, 405–426. https://doi.org/10.1088/0266-
5611/14/3/004
Mosegaard, K., Tarantola, A., 1995. Monte Carlo sampling of solutions to inverse
problems. J. Geophys. Res. (Solid Earth) 100, 12431–12448.
https://doi.org/10.1029/94JB03097
Moysey, S., Singha, K., Knight, R., 2005. A framework for inferring field-scale rock
physics relationships through numerical simulation. Geophys. Res. Lett. 32, 1–4.
https://doi.org/10.1029/2004GL022152
Nenna, V., Pidlisecky, A., Knight, R., 2011. Informed experimental design for electrical
resistivity imaging. Near Surf. Geophys. 9, 469–482. https://doi.org/10.3997/1873-
0604.2011027
References
105
Neuman, S.P., Xue, L., Ye, M., Lu, D., 2012. Bayesian analysis of data-worth
considering model and parameter uncertainties. Adv. Water Resour. 36, 75–85.
https://doi.org/10.1016/j.advwatres.2011.02.007
Nguyen, F., Garambois, S., Jongmans, D., Pirard, E., Loke, M.H., 2005. Image
processing of 2D resistivity data for imaging faults. J. Appl. Geophys. 57, 260–
277. https://doi.org/10.1016/j.jappgeo.2005.02.001
Nguyen, F., Kemna, A., Robert, T., Hermans, T., 2016. Data-driven selection of the
minimum-gradient support parameter in time-lapse focused electric imaging.
Geophysics 81, A1–A5. https://doi.org/10.1190/geo2015-0226.1
Nimmer, R.E., Osiensky, J.L., Binley, A.M., Williams, B.C., 2008. Three-dimensional
effects causing artifacts in two-dimensional, cross-borehole, electrical imaging. J.
Hydrol. 359, 59–70. https://doi.org/10.1016/j.jhydrol.2008.06.022
Nolet, G., Montelli, R., Virieux, J., 1999. Explicit, approximate expressions for the
resolution and a posteriori covariance of massive tomographic systems. Geophys.
J. Int. 138, 36–44. https://doi.org/10.1046/j.1365-246X.1999.00858.x
Nowak, W., Bode, F., Loschko, M., 2015. A Multi-objective Optimization Concept for
Risk-based Early-warning Monitoring Networks in Well Catchments. Procedia
Environ. Sci. 25, 191–198. https://doi.org/10.1016/j.proenv.2015.04.026
Nowak, W., Rubin, Y., de Barros, F.P.J., 2012. A hypothesis-driven approach to
optimize field campaigns. Water Resour. Res. 48, W06509.
https://doi.org/10.1029/2011WR011016
Oldenborger, G.A., Knoll, M.D., Routh, P.S., LaBrecque, D.J., 2007. Time-lapse ERT
monitoring of an injection/withdrawal experiment in a shallow unconfined
aquifer. Geophysics 72, F177–F187. https://doi.org/10.1190/1.2734365
Oldenborger, G.A., Routh, P.S., 2009. The point-spread function measure of resolution
for the 3-D electrical resistivity experiment. Geophys. J. Int. 176, 405–414.
https://doi.org/10.1111/j.1365-246X.2008.04003.x
Oldenborger, G.A., Routh, P.S., Knoll, M.D., 2005. Sensitivity of electrical resistivity
tomography data to electrode position errors. Geophys. J. Int. 163, 1–9.
https://doi.org/10.1111/j.1365-246X.2005.02714.x
Oldenburg, D.W., Li, Y., 1999. Estimating depth of investigation in dc resistivity and
IP surveys. Geophysics 64, 403. https://doi.org/10.1190/1.1444545
References
106
Oware, E.K., Moysey, S.M.J., Khan, T., 2013. Physically based regularization of
hydrogeophysical inverse problems for improved imaging of process-driven
systems. Water Resour. Res. 49, 6238–6247. https://doi.org/10.1002/wrcr.20462
Park, H., Scheidt, C., Fenwick, D., Boucher, A., Caers, J., 2013. History matching and
uncertainty quantification of facies models with multiple geological
interpretations. Comput. Geosci. 17, 609–621. https://doi.org/10.1007/s10596-013-
9343-5
Parsekian, A.D., Claes, N., Singha, K., Minsley, B.J., Carr, B., Voytek, E., Harmon, R.,
Kass, A., Carey, A., Thayer, D., Flinchum, B., 2017. Comparing Measurement
Response and Inverted Results of Electrical Resistivity Tomography Instruments.
J. Environ. Eng. Geophys. 22, 249–266. https://doi.org/10.2113/JEEG22.3.249
Parsekian, A.D., Singha, K., Minsley, B.J., Holbrook, W.S., Slater, L., 2015. Multiscale
geophysical imaging of the critical zone. Rev. Geophys. 53, 1–26.
https://doi.org/10.1002/2014RG000465
Pollock, D., Cirpka, O.A., 2010. Fully coupled hydrogeophysical inversion of synthetic
salt tracer experiments. Water Resour. Res. 46, W07501.
https://doi.org/10.1029/2009WR008575
Portniaguine, O., Zhdanov, M.S., 1999. Focusing geophysical inversion images.
GEOPHYSICS 64, 874–887. https://doi.org/10.1190/1.1444596
Ramirez, A.L., Daily, W.D., Newmark, R.L., 1995. Electrical resistance tomography for
steam injection monitoring and process control. J. Environ. Eng. Geophys. 1, 39.
https://doi.org/10.4133/JEEG1.A.39
Ramirez, A.L., Nitao, J.J., Hanley, W.G., Aines, R., Glaser, R.E., Sengupta, S.K., Dyer,
K.M., Hickling, T.L., Daily, W.D., 2005. Stochastic inversion of electrical
resistivity changes using a Markov Chain Monte Carlo approach. J. Geophys. Res.
110, B02101. https://doi.org/10.1029/2004JB003449
Ray, A., Kaplan, S., Washbourne, J., Albertin, U., 2018. Low frequency full waveform
seismic inversion within a tree based Bayesian framework. Geophys. J. Int. 212,
522–542. https://doi.org/10.1093/gji/ggx428
Ray, A., Sekar, A., Hoversten, G.M., Albertin, U., 2016. Frequency domain full
waveform elastic inversion of marine seismic data from the Alba field using a
Bayesian trans-dimensional algorithm. Geophys. J. Int. 205, 915–937.
https://doi.org/10.1093/gji/ggw061
References
107
Reichle, R.H., Crow, W.T., Keppenne, C.L., 2008. An adaptive ensemble Kalman filter
for soil moisture data assimilation. Water Resour. Res. 44, 1–13.
https://doi.org/10.1029/2007WR006357
Robinson, J., Johnson, T., Slater, L., 2015. Challenges and opportunities for fractured
rock imaging using 3D cross borehole electrical resistivity. Geophysics 80, E49–
E61. https://doi.org/10.1190/GEO2014-0138.1
Rojas, R., Feyen, L., Dassargues, A., 2008. Conceptual model uncertainty in
groundwater modeling: Combining generalized likelihood uncertainty
estimation and Bayesian model averaging. Water Resour. Res. 44, 1–16.
https://doi.org/10.1029/2008WR006908
Romero-Ruiz, A., Linde, N., Keller, T., Or, D., 2018. A Review of Geophysical Methods
for Soil Structure Characterization. Rev. Geophys. 56, 672–697.
https://doi.org/10.1029/2018RG000611
Ross, M., 2004. Consideration of available geophysical techniques for non-intrusive
survey of B291 trenches. Sellafield, United Kingdom.
Roubinet, D., Irving, J., 2014. Discrete-dual-porosity model for electric current flow in
fractured rock. J. Geophys. Res. Solid Earth 119, 767–786.
https://doi.org/10.1002/2013JB010668
Routh, P.S., 2009. A practical strategy to interrogate resolution , uncertainty and value
of information in geophysical inverse problem. Geohorizons 27–37.
Ruggeri, P., Gloaguen, E., Lefebvre, R., Irving, J., Holliger, K., 2014. Integration of
hydrological and geophysical data beyond the local scale: Application of
Bayesian sequential simulation to field data from the Saint-Lambert-de-Lauzon
site, Québec, Canada. J. Hydrol. 514, 271–280.
https://doi.org/10.1016/J.JHYDROL.2014.04.031
Ruggeri, P., Irving, J., Gloaguen, E., Holliger, K., 2013. Regional-scale integration of
multiresolution hydrological and geophysical data using a two-step Bayesian
sequential simulation approach. Geophys. J. Int. 194, 289–303.
https://doi.org/10.1093/gji/ggt067
Salamon, P., Feyen, L., 2010. Disentangling uncertainties in distributed hydrological
modeling using multiplicative error models and sequential data assimilation.
Water Resour. Res. 46, 1–20. https://doi.org/10.1029/2009WR009022
Scales, J.A., Snieder, R., Treitel, S., 1997. To Bayes or not to Bayes ? Geophysics 62, 1045–
1046.
References
108
Scheidt, C., Li, L., Caers, J., 2018. Quantifying Uncertainty in Subsurface Systems,
Geophysical Monograph Series. John Wiley & Sons, Inc., Hoboken, NJ, USA.
https://doi.org/10.1002/9781119325888
Schmidt-Hattenberger, C., Bergmann, P., Labitzke, T., Wagner, F., Rippe, D., 2016.
Permanent crosshole electrical resistivity tomography (ERT) as an established
method for the long-term CO2 monitoring at the Ketzin pilot site. Int. J. Greenh.
Gas Control 52, 432–448. https://doi.org/10.1016/j.ijggc.2016.07.024
Schöniger, A., Nowak, W., Hendricks Franssen, H.-J., 2012. Parameter estimation by
ensemble Kalman filters with transformed data: Approach and application to
hydraulic tomography. Water Resour. Res. 48, W04502.
https://doi.org/10.1029/2011WR010462
Shi, L., Zeng, L., Tang, Y., Chen, C., Yang, J., 2013. Uncertainty quantification of
contaminant transport and risk assessment with conditional stochastic
collocation method. Stoch. Environ. Res. Risk Assess. 27, 1453–1464.
https://doi.org/10.1007/s00477-012-0682-x
Shi, W., 1998. Advanced modeling and inversion techniques for three-dimensional
geoelectrical surveys. Massachusetts Institute of Technology.
Sibbertt, L., Chambers, J.E., Li, B., Wilkinson, P.B., 2017. Total Generalised Variation:
An improved regulariser for Electrical Resistivity Tomography inversion., in:
Geophysical Research Abstracts. pp. EGU2017-14348.
Singha, K., Day-Lewis, F.D., Johnson, T., Slater, L.D., 2015. Advances in interpretation
of subsurface processes with time-lapse electrical imaging. Hydrol. Process. 29,
1549–1576. https://doi.org/10.1002/hyp.10280
Singha, K., Gorelick, S.M., 2006. Effects of spatially variable resolution on field-scale
estimates of tracer concentration from electrical inversions using Archie’s law.
Geophysics 71, G83. https://doi.org/10.1190/1.2194900
Skyes, J.F., Wilson, J.L., Andrews, R.W., 1985. Sensitivity analysis for steady state
groundwater flow using adjoint operators. Water Resour. Res. 21, 359–371.
Slater, L., Binley, A., 2003. Evaluation of permeable reactive barrier (PRB) integrity
using electrical imaging methods. Geophysics 68, 911–921.
https://doi.org/10.1190/1.1581043
Slater, L., Binley, A.M., Daily, W., Johnson, R., 2000. Cross-hole electrical imaging of a
controlled saline tracer injection. J. Appl. Geophys. 44, 85–102.
https://doi.org/10.1016/S0926-9851(00)00002-1
References
109
Slater, L.D., Binley, A., Brown, D., 1997. Electrical Imaging of Fractures Using Ground-
Water Salinity Change. Ground Water 35, 436–442. https://doi.org/10.1111/j.1745-
6584.1997.tb00103.x
Sobol′, I.., 2001. Global sensitivity indices for nonlinear mathematical models and their
Monte Carlo estimates. Math. Comput. Simul. 55, 271 – 280.
https://doi.org/10.1016/S0378-4754(00)00270-6
Stummer, P., Maurer, H.R., Green, A.G., 2004. Experimental design: Electrical
resistivity data sets that provide optimum subsurface information. Geophysics
69, 120–139.
Tarantola, A., 2006. Popper, Bayes and the inverse problem. Nat. Phys. 2, 492–494.
https://doi.org/10.1038/nphys375
Tarantola, A., 2005. Inverse Problem Theory and Methods for Model Parameter
Estimation. SIAM.
Trainor-Guitton, W.J., Ramirez, A., Yang, X., Mansoor, K., Sun, Y., Carroll, S., 2013.
Value of information methodology for assessing the ability of electrical
resistivity to detect CO2/brine leakage into a shallow aquifer. Int. J. Greenh. Gas
Control 18, 101–113. https://doi.org/10.1016/j.ijggc.2013.06.018
Tran, A.P., Dafflon, B., Hubbard, S.S., 2017. Coupled land surface–subsurface
hydrogeophysical inverse modeling to estimate soil organic carbon content and
explore associated hydrological and thermal dynamics in the Arctic tundra.
Cryosph. 11, 2089–2109. https://doi.org/10.5194/tc-11-2089-2017
Tran, A.P., Dafflon, B., Hubbard, S.S., Kowalsky, M.B., Long, P., Tokunaga, T.K.,
Williams, K.H., 2016. Quantifying shallow subsurface water and heat dynamics
using coupled hydrological-thermal-geophysical inversion. Hydrol. Earth Syst.
Sci. 20, 3477–3491. https://doi.org/10.5194/hess-20-3477-2016
Trigo, F.C., Gonzalez-Lima, R., Amato, M.B.P., 2004. Electrical impedance tomography
using the extended Kalman filter. IEEE Trans. Biomed. Eng. 51, 72–81.
https://doi.org/10.1109/TBME.2003.820389
Truex, M.J., Johnson, T.C., Strickland, C.E., Peterson, J.E., Hubbard, S.S., 2013.
Monitoring Vadose Zone Desiccation with Geophysical Methods. Vadose Zo. J.
12, 0. https://doi.org/10.2136/vzj2012.0147
Tsai, F.T.C., Elshall, A., 2013. Hierarchical Bayesian model averaging for
hydrostratigraphic modeling: Uncertainty segregation and comparative
evaluation. Water Resour. Res. 49, 5520–5536. https://doi.org/10.1002/wrcr.20428
References
110
Tso, C.-H.M., Kuras, O., Wilkinson, P.B., Uhlemann, S., Chambers, J.E., Meldrum, P.I.,
Graham, J., Sherlock, E.F., Binley, A., 2017. Improved characterisation and
modelling of measurement errors in electrical resistivity tomography (ERT)
surveys. J. Appl. Geophys. 146, 103–119.
https://doi.org/10.1016/J.JAPPGEO.2017.09.009
Tso, C.M., Kuras, O., Binley, A., 2019. On the field estimation of moisture content using
electrical geophysics ‐ the impact of petrophysical model uncertainty. Water
Resour. Res. 2019WR024964. https://doi.org/10.1029/2019WR024964
Uhlemann, S., Wilkinson, P.B., Maurer, H., Wagner, F.M., Johnson, T.C., Chambers,
J.E., 2018. Optimized survey design for Electrical Resistivity Tomography:
combined optimization of measurement configuration and electrode placement.
Geophys. J. Int.
Vauhkonen, M., Karjalainen, P.A., Kaipio, J.P., 1998. A Kalman filter approach to track
fast impedance changes in electrical impedance tomography. IEEE Trans Biomed
Eng 45, 486–493. https://doi.org/10.1109/10.664204
Vrugt, J.A., Robinson, B.A., 2007. Improved evolutionary optimization from
genetically adaptive multimethod search. Proc. Natl. Acad. Sci. 104, 708–711.
https://doi.org/10.1073/pnas.0610471104
Vrugt, J.A., Robinson, B.A., Hyman, J.M., 2009. Self-Adaptive Multimethod Search for
Global Optimization in Real-Parameter Spaces. IEEE Trans. Evol. Comput. 13,
243–259. https://doi.org/10.1109/TEVC.2008.924428
Wagner, F.M., Bergmann, P., Rücker, C., Wiese, B., Labitzke, T., Schmidt-Hattenberger,
C., Maurer, H., 2015a. Impact and mitigation of borehole related effects in
permanent crosshole resistivity imaging: An example from the Ketzin CO2
storage site. J. Appl. Geophys. 123, 102–111.
https://doi.org/10.1016/j.jappgeo.2015.10.005
Wagner, F.M., Günther, T., Schmidt-Hattenberger, C., Maurer, H., 2015b. Constructive
optimization of electrode locations for target-focused resistivity monitoring.
Geophysics 80, E29–E40. https://doi.org/10.1190/geo2014-0214.1
Wagner, F.M., Wiese, B.U., 2018. Fully coupled inversion on a multi-physical reservoir
model – Part II: The Ketzin CO2 storage reservoir. Int. J. Greenh. Gas Control 75,
273–281. https://doi.org/10.1016/J.IJGGC.2018.04.009
Wainwright, H.M., Finsterle, S., Jung, Y., Zhou, Q., Birkholzer, J.T., 2014. Making sense
of global sensitivity analyses. Comput. Geosci. 65, 84–94.
https://doi.org/10.1016/j.cageo.2013.06.006
References
111
Wainwright, H.M., Flores Orozco, A., Bücker, M., Dafflon, B., Chen, J., Hubbard, S.S.,
Williams, K.H., 2016. Hierarchical Bayesian method for mapping biogeochemical
hot spots using induced polarization imaging. Water Resour. Res. 52, 533–551.
https://doi.org/10.1002/2015WR017763
Ward, W.O.C., Wilkinson, P.B., Chambers, J.E., Oxby, L.S., Bai, L., 2014. Distribution-
based fuzzy clustering of electrical resistivity tomography images for interface
detection. Geophys. J. Int. 197, 310–321. https://doi.org/10.1093/gji/ggu006
Wehrer, M., Slater, L.D., 2015. Characterization of water content dynamics and tracer
breakthrough by 3-D electrical resistivity tomography (ERT) under transient
unsaturated conditions. Water Resour. Res. 51, 97–124.
https://doi.org/10.1002/2014WR016131
Whiteley, J.S., Chambers, J.E., Uhlemann, S., Wilkinson, P.B., Kendall, J.M., 2018.
Geophysical monitoring of moisture-induced landslides: a review. Rev. Geophys.
https://doi.org/10.1029/2018RG000603
Wiese, B., Wagner, F., Norden, B., Maurer, H., Schmidt-Hattenberger, C., 2018. Fully
coupled inversion on a multi-physical reservoir model - Part I: Theory and
concept. Int. J. Greenh. Gas Control.
Wilkinson, P., Chambers, J., Uhlemann, S., Meldrum, P., Smith, A., Dixon, N., Loke,
M.H., 2016. Reconstruction of landslide movements by inversion of 4-D electrical
resistivity tomography monitoring data. Geophys. Res. Lett. 43, 1166–1174.
https://doi.org/10.1002/2015GL067494
Wilkinson, P.B., Chambers, J.E., Lelliott, M., Wealthall, G.P., Ogilvy, R.D., 2008.
Extreme sensitivity of crosshole electrical resistivity tomography measurements
to geometric errors. Geophys. J. Int. 173, 49–62. https://doi.org/10.1111/j.1365-
246X.2008.03725.x
Wilkinson, P.B., Chambers, J.E., Meldrum, P.I., Gunn, D.A., Ogilvy, R.D., Kuras, O.,
2010a. Predicting the movements of permanently installed electrodes on an
active landslide using time-lapse geoelectrical resistivity data only. Geophys. J.
Int. 183, 543–556. https://doi.org/10.1111/j.1365-246X.2010.04760.x
Wilkinson, P.B., Loke, M.H., Meldrum, P.I., Chambers, J.E., Kuras, O., Gunn, D.A.,
Ogilvy, R.D., 2012. Practical aspects of applied optimized survey design for
electrical resistivity tomography. Geophys. J. Int. 189, 428–440.
https://doi.org/10.1111/j.1365-246X.2012.05372.x
Wilkinson, P.B., Meldrum, P.I., Chambers, J.E., Kuras, O., Ogilvy, R.D., 2006. Improved
strategies for the automatic selection of optimized sets of electrical resistivity
References
112
tomography measurement configurations. Geophys. J. Int. 167, 1119–1126.
https://doi.org/10.1111/j.1365-246X.2006.03196.x
Wilkinson, P.B., Meldrum, P.I., Kuras, O., Chambers, J.E., Holyoake, S.J., Ogilvy, R.D.,
2010b. High-resolution Electrical Resistivity Tomography monitoring of a tracer
test in a confined aquifer. J. Appl. Geophys. 70, 268–276.
https://doi.org/10.1016/j.jappgeo.2009.08.001
Wilkinson, P.B., Uhlemann, S., Meldrum, P.I., Chambers, J.E., Carrière, S., Oxby, L.S.,
Loke, M.H.H., 2015. Adaptive time-lapse optimized survey design for electrical
resistivity tomography monitoring. Geophys. J. Int. 203, 755–766.
https://doi.org/10.1093/gji/ggv329
Xue, L., Zhang, D., 2014. A multimodel data assimilation framework via the ensemble
Kalman filter. Water Resour. Res. 50, 4197–4219.
https://doi.org/10.1002/2013WR014525
Yang, X., Chen, X., Carrigan, C.R., Ramirez, A.L., 2014. Uncertainty quantification of
CO2 saturation estimated from electrical resistance tomography data at the
Cranfield site. Int. J. Greenh. Gas Control 27, 59–68.
https://doi.org/10.1016/j.ijggc.2014.05.006
Yang, X., Lassen, R.N., Jensen, K.H., Looms, M.C., 2015. Monitoring CO2 migration in
a shallow sand aquifer using 3D crosshole electrical resistivity tomography. Int.
J. Greenh. Gas Control 42, 534–544. https://doi.org/10.1016/J.IJGGC.2015.09.005
Yeh, T.-C.J., Liu, S., 2000. Hydraulic tomography : Development of a new aquifer test
method. Water Resour. Res. 36, 2095–2105.
Yeh, T.-C.J., Liu, S., Glass, R.J., Baker, K., Brainard, J.R., Alumbaugh, D., LaBrecque, D.,
2002. A geostatistically based inverse model for electrical resistivity surveys and
its applications to vadose zone hydrology. Water Resour. Res. 38, WR001204.
https://doi.org/10.1029/2001WR001204
Zhou, J., Revil, A., Jardani, A., 2016. Stochastic structure-constrained image-guided
inversion of geophysical data. Geophysics 81, E89–E101.
https://doi.org/10.1190/geo2014-0569.1
Zhou, J., Revil, A., Karaoulis, M., Hale, D., Doetsch, J., Cuttler, S., 2014. Image-guided
inversion of electrical resistivity data. Geophys. J. Int. 197, 292–309.
https://doi.org/10.1093/gji/ggu001
Appendix 1: Derivation of the Gauss-Newton solution to the regularized inverse
problem
113
Appendix 1: Derivation of the Gauss-Newton solution to the
regularized inverse problem
We begin by writing the objective function 𝜙 = 𝜙(m) using Taylor expansion and set
m = m𝑘 + Δm
𝜙(m𝑘 + Δm) = 𝜙(m𝑘) + 𝛁𝜙(m𝑘)Δm+1
2ΔmT𝛁2𝜙(m𝑘)Δm + 𝐻.𝑂. 𝑇. (11)
where m𝑘 is model estimate at the 𝑘-th iteration, Δm is model increment, and H.O.T.
denotes higher-order terms. At the true solution m = m∗, 𝜙 reaches a minimum so the
derivative of the above equation is zero. Therefore, we differentiate both sides of the
equation above, drop the higher-order terms (third derivatives or above) and set it = 0
𝛁𝜙(m𝑘 + Δm) ≈ 𝛁𝜙(m𝑘) + 𝛁2𝜙(m𝑘)Δm = 0 (12)
Rearranging the terms give:
𝛁2𝜙(𝐦𝑘)𝚫𝐦 = −𝛁𝜙(𝐦𝑘) (13)
Now we seek to obtain the first and second derivatives of the objective function. The
commonly used objective function takes the form of 𝜙 = 𝜙𝑑 + α𝜙𝑚 , where α is a
scalar damping factor, 𝜙𝑑 and 𝜙𝑚 are the data misfit and model roughness objective
functions, respectively.
The data misfit function 𝜙𝑑 is the sum of data residuals 𝑟𝑖 squared (or equivalently in
matrix notation):
𝜙𝑑 = ∑ 𝑟𝑖2𝑁
𝑖=1 = ∑ ( 𝑑𝑖−𝐹𝑖(m)
𝜎𝑖)2
𝑁𝑖=1 or 𝜙𝑑 = (d − F(m))
𝑇𝐖𝑑𝑇𝐖𝑑(d − F(m)) (14)
where the data weighting matrix 𝐖𝑑 is a diagonal matrix and its (𝑖, 𝑖)-th entry is given
by 1/𝜎𝑖 . The model roughness objective function 𝜙𝑚 is (or equivalently in matrix
notation):
𝜙𝑚 = ∑ ∑ (𝑚𝑗 −𝑚𝑗′)2
𝑗′∈𝑆𝑗𝑀𝑗=1 or 𝜙𝑚 = m
𝑘𝑇𝐑𝒎m𝑘 (15)
where 𝑆𝑗 is a set of parameter indices that are neighbors of 𝑗. The roughness matrix 𝐑
is a symmetric matrix with integer coefficient. The sum of coefficients in each row and
each column should be zero.
Appendix 1: Derivation of the Gauss-Newton solution to the regularized inverse
problem
114
Therefore, the combined objective function is:
𝜙 = 𝜙𝑑 + α𝜙𝑚 =∑( 𝑑𝑖 − 𝐹𝑖(m)
𝜎𝑖)2𝑁
𝑖=1
+ α∑∑(𝑚𝑗 −𝑚𝑗′)2
𝑀′
𝑗′=1
𝑀
𝑗=1
or 𝜙⏟1×1
= 𝜙𝑑 + α𝜙𝑚 = (d − F(m))𝑇𝐖𝑑𝑇 𝐖𝑑⏟𝑁×𝑁
(d − F(m))⏟ 𝑁×1
+ αm𝑘𝑇𝐑𝒎⏟𝑀×𝑀
m𝑘⏟𝑀×1
(16)
Taking derivative with respect to m gives the first derivative of 𝜙,
𝜕𝜙
𝜕𝑚𝑗= −2∑(
𝑑𝑖 − 𝐹𝑖(m)
𝜎𝑖)( 1
𝜎𝑖)𝜕𝐹𝑖𝜕𝑚𝑗
+ 2α∑∑(𝑚𝑗 −𝑚𝑗′)
𝑀′
𝑗′=1
𝑀
𝑗=1
𝑁
𝑖=1
or 𝛁𝜙(m𝑘) = 2[−𝐉𝑇𝐖𝑑𝑇𝐖𝑑(d − F(m)) + 𝛼𝐑𝒎m
𝑘] (17)
where [𝜕𝐹1/𝜕𝑚1 ⋯ 𝜕𝐹1/𝜕𝑚𝑀
⋮ ⋱ ⋮𝜕𝐹𝑁/𝜕𝑚1 ⋯ 𝜕𝐹𝑁/𝜕𝑚𝑀
] are the entries of the Jacobian matrix J of size 𝑁 ×𝑀.
We take derivative with respect to 𝑚 again to obtain the second derivative of 𝜙,
𝜕2𝜙
𝜕𝑚𝑗2= −2∑(
𝑑𝑖 − 𝐹𝑖(m)
𝜎𝑖)( 1
𝜎𝑖)𝜕2𝐹𝑖𝜕𝑚𝑗
2
𝑁
𝑖=1
+ 2∑𝜕𝐹𝑖𝜕𝑚𝑗
( 1
𝜎𝑖)2 𝜕𝐹𝑖𝜕𝑚𝑗
𝑁
𝑖=1
+ 2α∑∑(𝑗 − 𝑗′)
𝑀′
𝑗′=1
𝑀
𝑗=1
for j = 1,… ,m
or 𝛁2𝜙(m𝑘) = 2[−𝛁𝐉𝑇𝐖𝑑𝑇𝐖𝑑(d − F(m)) + 𝐉
𝑇𝐖𝑑𝑇𝐖𝑑𝐉 + 𝛼𝐑𝒎]
According to Oldenburgh and Li (2005), the first terms accounts for the change in the
sensitivity when the model is perturbed. It is computationally burdensome to compute
and also its importance decreases as the optimization proceeds because the difference
between the predicted and observed data become smaller. For these reasons this term
is generally neglected and we use an approximate Hessian in the computations:
𝜕2𝜙
𝜕𝑚𝑗2≈ 2∑
𝜕𝐹𝑖𝜕𝑚𝑗
( 1
𝜎𝑖)2 𝜕𝐹𝑖𝜕𝑚𝑗
𝑁
𝑖=1
+ 2α∑∑(𝑗 − 𝑗′)
𝑀′
𝑗′=1
𝑀
𝑗=1
for j = 1,… ,m
or 𝛁2𝜙(m𝑘) ≈ 2[𝐉𝑇𝐖𝑑𝑇𝐖𝑑𝐉 + 𝛼𝐑𝒎] (18)
The approximate Hessian is positive semi-definite so its inverse exists. Finally, by
combining equations (13), (17) and (18) into:
𝟐[𝐉𝑇𝐖𝑑𝑇𝐖𝑑𝐉 + 𝛼𝐑𝒎]Δm = −𝟐[−𝐉
𝑇𝐖𝑑𝑇𝐖𝑑(d − F(m)) + 𝛼𝐑𝒎m
𝑘] (19)
Cancelling and rearranging terms:
(𝐉𝑇𝐖𝑑𝑇𝐖𝑑𝐉 + 𝛼𝐑𝒎)⏟ 𝑀×𝑀
Δm⏟𝑀×1
= 𝐉𝑇𝐖𝑑𝑇𝐖𝑑(d − F(m)) − 𝛼𝐑𝒎m
𝑘⏟
𝑀×1
(20)
Paper 1: Improved characterisation and modelling of measurement errors in electrical
resistivity tomography (ERT) surveys
115
4. Paper 1: Improved characterisation and modelling of
measurement errors in electrical resistivity tomography (ERT)
surveys
Published as Tso, C.-H.M., Kuras, O., Wilkinson, P., Uhlemann, S., Chambers, J.,
Meldrum, P., Graham, J., Sherlock, E., Binley, A. (2017): Improved modelling and
characterisation of measurement errors in electrical resistivity tomography (ERT)
surveys. Journal of Applied Geophysics, 145, 103--119,
DOI:10.1016/j.jappgeo.2017.09.009.
Abstract
116
Abstract
Measurement errors can play a pivotal role in geophysical inversion. Most
inverse models require users to prescribe or assume a statistical model of data errors
before inversion. Wrongly prescribed errors can lead to over- or under-fitting of data,
however, the derivation of models of data errors is often neglected. With the
heightening interest in uncertainty estimation within hydrogeophysics, better
characterisation and treatment of measurement errors is needed to provide improved
image appraisal. Here we focus on the role of measurement errors in electrical
resistivity tomography (ERT). We have analysed two time-lapse ERT datasets: one
contains 96 sets of direct and reciprocal data collected from a surface ERT line within
a 24h timeframe; the other is a two-year-long cross-borehole survey at a UK nuclear
site with 246 sets of over 50,000 measurements. Our study includes the characterisation
of the spatial and temporal behaviour of measurement errors using autocorrelation
and correlation coefficient analysis. We find that, in addition to well-known
proportionality effects, ERT measurements can also be sensitive to the combination of
electrodes used, i.e. errors may not be uncorrelated as often assumed. Based on these
findings, we develop a new error model that allows grouping based on electrode
number in addition to fitting a linear model to transfer resistance. The new model
explains the observed measurement errors better and shows superior inversion results
and uncertainty estimates in synthetic examples. It is robust, because it groups errors
together based on the electrodes used to make the measurements. The new model can
be readily applied to the diagonal data weighting matrix widely used in common
inversion methods, as well as to the data covariance matrix in a Bayesian inversion
framework. We demonstrate its application using extensive ERT monitoring datasets
from the two aforementioned sites.
Graphical abstract
117
Graphical abstract
Probability density functions (PDF) of different ERT errors for 24h of surface ERT data collected at a
wetland site in the UK. The mean repeatability errors generally increase with the period of time
considered. Reciprocal errors generally agree with short-term repeatability errors. The PDF of stacking
errors shows much lower mean and variance. Using stacking errors as a measure of measurement errors
may lead to overfitting of data during inversion and underestimation of uncertainty.
Highlights
Stacking, reciprocal and repeatability errors are compared using statistical
analysis
Having common electrodes increase correlation between measurements
A new error model based on grouping the electrodes used is developed
The new model yields better inversion results and uncertainty estimates
Keywords
ERT, resistivity, measurement errors, uncertainty, linear mixed effects, inversion
-7 -6 -5 -4 -3 -2 -1 0 1 20
0.1
0.2
0.3
0.4
0.5
log10
(errors)
Pro
babili
ty d
ensity
PDF of repeatability and reciprocal errors at Sellafield
stacking
reciprocal
14d repeatability
-6 -5.5 -5 -4.5 -4 -3.5 -3 -2.5 -20
1
2
3
4
5
log10
(errors)
Pro
babili
ty d
ensity
PDF of repeatability and reciprocal errors at Boxford
stacking errors
1/2h repeatability
1h repeatability
2hrepeatability
24h repeatability
reciprocal
Introduction
118
1) Introduction
Measurement errors are an integral part of scientific observations. Properly
describing such errors is essential to harness the information about the observed
behaviour contained in the measurements. Measurement errors may be random or
systematic. In commonly used geophysical inverse methods, measurement errors are
assumed to be uncorrelated and random. In the context of an inversion, the total data
error is given by the square root of the sum of squares of measurement errors and
modelling errors. Sources of modelling errors include inaccuracy of the forward model
(e.g. due to discretisation of a numerical model) and appropriateness of a forward
model (e.g. representing a 3D problem using a 2D model). Numerical modelling errors
are relatively well understood because they can be studied by comparing forward
modelling data of a homogeneous domain with analytical solutions (see Binley, 2015).
We, therefore, focus here on measurement errors, in particular within the context of
electrical resistivity tomography (ERT).
1.1. The role of ERT measurement errors
Measurement error estimates play a critical role in ERT inversion (see more in
section 2.3). They affect the amount of damping imposed on the data and also the point
at which convergence is attained. Both of the above are achieved by weighting data in
the objective function, and thus, measurement error estimates control whether there
will be over-fitting or under-fitting of data during inversion. This concept can be
illustrated by comparing various inverted images. Figure 1 shows the results of
inverting synthetic ERT experiments corrupted by 5% Gaussian noise. In the synthetic
domain, a resistive target is placed between x = 15 m and x = 20 m and the topsoil is
relatively conductive (Figure 1a). Inverting the data with 10% assumed Gaussian noise
leads to under-fitting and a very smooth resultant image (Figure 1b), while assuming
2% noise leads to over-fitting and a number of artefacts (Figure 1d). This simple
example shows that inversion results can be sensitive to the assumed measurement
error levels. Failure to prescribe them adequately can significantly change the resultant
image.
Introduction
119
Attempts have been made to account for data errors in a more sophisticated
manner. Robust inversion (Kemna, 2000; Morelli and LaBrecque, 1996) adjusts error
weights when there are apparent outliers. It is important, however, to notice that the
outliers are linked to a specific error weight derived a priori by the error model—they
may not be outliers anymore if a different error model is used. Similarly, in Bayesian
inversion (e.g. Irving and Singha, 2010; Ramirez et al., 2005), one needs to prescribe the
estimated data uncertainty in the likelihood function. While different inversion
strategies handle measurement errors differently, a robust and accurate prescription
of measurement errors is essential to obtain reliable and realistic inversion results.
The impact of measurement errors is not limited to inversion—it is a natural
extension of stochastic inversion where posterior models are estimates of uncertainty,
whereas deterministic inversion results (or an ensemble of them) can further be used
to estimate uncertainty via Monte Carlo approaches. Therefore, uncertainties in
measurements would propagate to uncertainties of model estimates. Similarly, if the
inversion results are used to detect or monitor subsurface processes, or to infer
hydrological properties, their associated errors can be traced back to measurement
errors. It is apparent that measurement errors propagate through the various stages of
a hydrogeophysics study workflow. With the heightening interest in uncertainty
estimation within hydrogeophysics (Binley et al., 2015; Huisman et al., 2010; Linde et
al., 2015b; Rubin and Hubbard, 2005; Vereecken et al., 2006), better characterisation
and treatment of measurement errors is necessary to provide better image appraisal.
1.2. Measurement errors in ERT: a review
The handling of measurement errors in ERT surveys, despite its importance as
outlined above, is variable. The simplest (but not necessarily the most reliable) method
of assessing a measurement error in an ERT measurement is through the use of
stacking, i.e. the repeated measurement of transfer resistance through a number of
cycles of current injection. Such stacking assessment offers valuable in-field data
quality appraisal but, as shown later, may be of limited value in quantifying a data
weight for ERT inversion. Alternatively, repeatability errors can be obtained by
multiple, separate measurements of transfer resistance over time. Usually this involves
Introduction
120
a repetition of the entire ERT measurement sequence sometime after the first attempt.
Reciprocity checks are another method of measurement error assessment. Reciprocity
is the general physical principle where the switching of source/sink and observation
locations would yield the same response (Parasnis, 1988). It is, for example, utilised in
groundwater hydrology (Barker, 1991; Bruggeman, 1972; Delay et al., 2011; Falade,
1981). Reciprocity checks for ERT are conducted by swapping the current and potential
electrodes. Reciprocity breaks down when the ground response is non-linear (i.e. non-
ohmic for ERT) or time-dependent (i.e. something changes between forward and
reciprocal measurements).
As LaBrecque et al. (1996a) point out, both repeated and reciprocal
measurements are measures of precision not accuracy. Sources of systematic error are
not accounted for explicitly in measurements of precision – some procedures may miss
them entirely while others lump them as random errors. Reciprocal errors treat the
swapping of electrodes as a way to account for some systematic errors while
repeatability errors do not consider them at all. Therefore, reciprocal errors may be
more useful to eliminate bias caused by using a particular pair of electrodes as
transmitter and another as receiver. The most commonly used errors in ERT, however,
are stacking errors and they are misreferred as repeatability errors (Day-Lewis et al.,
2008). Modern ERT instruments are equipped with stacking capability and they
automatically return stacking errors. In other words, stacking errors can be obtained
without re-running the measurement procedure, which is very attractive in time-
sensitive or time-consuming surveys.
We surveyed a number of published ERT studies and report their description
of error analysis in Table 1. From Table 1, we see that reciprocity is a commonly used
measure, while a small fraction of field and experimental studies do not report their
treatment of errors at all. Studies often attribute their exclusion of reciprocal errors to
logistical constraints and argue that reporting stacking errors is sufficient. After errors
are obtained, an error model (usually a linear relationship linking error to transfer
resistance) is established (Binley et al., 1995). Once obtained, such a relationship may
be used to predict the errors of individual measurements and thus contribute to the
Introduction
121
data weight in the inverse modelling. Some authors, however, assign observed errors
directly in the inverse modelling, although this is potentially flawed unless statistical
robustness of the quantified error is established (recognising that for most surveys
errors are only computed from two observations). This practice also makes it
impossible to identify “disinformative data” (Beven and Westerberg, 2011). From the
reported error models, it is observed that error levels are generally higher for cross-
borehole surveys, largely due to more challenging electrode contact conditions
(compared to most surface ERT array surveys). Prior to fitting the error model and
carrying out inversion, measurements with high errors are often eliminated;
sometimes more than 20% of the collected data are removed. For time-lapse studies, it
is quite common that the entire time series of an individual resistance measurement is
removed if any part of the time series is deemed to be an outlier. For recent work on
time-lapse cross-borehole ERT, see Schmidt-Hattenberger et al. (2016) and Yang et al.
(2014).
Error models are generally a function of average measured transfer resistance
(i.e. the error in a transfer resistance increases with the magnitude of transfer resistance)
because of the well-known proportionality effects (Aster et al., 2005) in DC resistivity
measurement errors (Binley, 2015; Binley et al., 1995). In studies where errors are
accounted for, there is generally a preference to use model-predicted errors rather than
individually observed errors since error assessment based on two observations is
potentially unreliable. Some studies mitigate this potential issue by binning (or
grouping) data with similar transfer resistance together before fitting an error model
(Koestel et al., 2008; Robinson et al., 2015; Wehrer and Slater, 2015). This practice
should give more robust error estimates, although the error model may vary with the
number of bins used.
To better characterise measurement errors, more understanding of the factors that
contribute to them is needed. Current practice leaves many of the assumptions in
ERT measurement errors modelling unchallenged. For example, do measurement
errors show temporal or spatial correlations? Can we improve from using linear
measurement error models? Are stacking errors and reciprocal errors comparable
Introduction
122
indicators of measurement errors? ERT surveys typically use each of the electrodes
for multiple measurements. Ramirez et al. (2005) notes that this may increase the
probability that measurement errors are correlated, however, there has been no
published work addressing this issue.
Figure 1 Synthetic problem for demonstration (a) Synthetic domain with a more conductive layer near
the surface and a resistive area between x = 15m and x = 20m. The synthetic data from running a forward
model in (a) is perturbed with 5% Gaussian noise and then inverted by assuming (b) 10% linear error
model (c) 5% linear error model (d) 2% linear error model. Note that rms error is defined as
√∑ (𝒐𝒃𝒔 − 𝒔𝒊𝒎)𝟐𝒏𝒊=𝟏 /𝒏 , where obs and sim are vectors of observed/true and simulated transferred
resistances of length n respectively. Note that the convergence target for all the inversions is a chi-
squared statistic of 1.
1.3. Recent work on ERT measurement errors
Attempts have been made to handle potential systematic effects of
measurement errors. Zhou and Dahlin (2003) studied the effect of spacing errors for 8
types of common 2D resistivity arrays. They confirm the common observation that ERT
error outliers are often correlated with high contact resistances for some of the
electrodes used in a measurement. Wilkinson et al. (2008) developed an approach to
Introduction
123
filter out configurations that are highly sensitive to geometric error in crosshole ERT
surveys. Similarly, Wilkinson et al. (2016, 2010a) developed techniques to recover
movements of permanently installed electrodes so that active landslides can be
monitored using time-lapse ERT data only. As the popularity of time-lapse surveys
increases, specific methods to handle and characterize measurement errors in large
time-lapse datasets emerge. Deceuster et al. (2013) developed a method to automate
the identification of changes in electrode contact during time-lapse ERT experiments.
More recently, Mitchell and Oldenburg (2016) developed a 4-step data quality control
methodology for very large ERT datasets.
Recently, Kim et al. (2016) proposed a new measurement protocol in which self-
potential (SP) data are obtained immediately prior to measuring DC. It involves
swapping the polarities of the two current electrodes in each measurement to obtain a
positive and a negative potential (i.e. thus a forward and backward resistance) for each
measurement. This protocol claims to account for SP effects in DC measurements and
eliminate distortions in the DC resistivity potential field caused by all unknown
mechanisms including ambient noise.
1.4. Outline of this work
This paper addresses a number of practical issues related to the treatment of
measurement errors in ERT inversion. We compare stacking, repeatability, and
reciprocal errors in their utility to describe errors in measurements. We also study
whether measurement errors are correlated in time and/or in space. We then
hypothesize that measurement errors in ERT are not only linearly dependent on
transfer resistances, but that the electrodes used in taking each measurement can be
used as a grouping variable to improve error characterisation. We show that using the
new error model leads to better inversion results and uncertainty estimates through
synthetic and field experiments data. We first describe the datasets and analysis
methods in section 2.1 and 2.2. Then we describe the ERT inversion and uncertainty
estimation methods used in section 2.3 and 2.4. Section 3) reports results for the error
analysis. We introduce a new error model based on linear mixed effects models and
grouping of electrodes in section 4), and section 5) shows results of inversion and
Approach
124
uncertainty quantification. We then discuss the implications of the results in section 6),
and provide conclusions and recommendations in section 7).
2) Approach
With recent advances made in the development of automated ERT systems,
ERT experiments can be conducted remotely, allowing the collection of a large volume
of background ERT measurements for quality control purposes. These rich datasets
can be exploited to investigate the behaviour of measurement errors through statistical
analysis. They provide opportunities to explore errors in ERT datasets, including the
assessment of temporal and spatial correlation of errors. We scrutinize two large field
datasets through statistical analysis of different types of measurement errors.
First, we examine the probability density functions for each error type, namely
stacking, reciprocal and repeatability errors. This allows us to understand the mean
and variance of their distributions. Next, we use autocorrelation and correlation
coefficient analysis to study the sequential and spatial correlation of errors between
measurements. Insights about the potential correlation in measurement errors can help
in developing improved error models. We study the validity of repeatability errors by
computing the autocorrelation and correlation coefficient of the departure from the
mean of repeated measurements instead of using the repeatability errors directly. If
repeatability errors are purely random, using any subset of the set of repeats for each
given measurement should give the same errors and thus the departure from the mean
should exhibit little correlation. We compare inversion of ERT data using different
error types and models on identical datasets to illustrate how they manifest in
inversion results. Lastly, we obtain uncertainty estimates of inversion results using a
Monte Carlo simulation procedure. This allows us to visualize how measurement
errors propagate into uncertainty in model estimates (in this study we assume there
are no other error sources).
Approach
125
Table 1 Table of error models reported in the literature
Approach
126
Approach
127
2.1. Dataset description
A synthetic dataset, along with two field datasets collected by the British
Geological Survey (BGS) are used for this work.
2.1.1. Synthetic Dataset
A synthetic dataset was created for use as an illustrative example using the
synthetic domain and array of Figure 1. The array consists of 25 2-m spaced surface
electrodes. As seen in Figure1(a), the resistivity structure of the domain consists of a
1m thick, 100 Ωm top layer. Beneath it is a 200 Ωm formation, in which a 500 Ωm unit
protrudes vertically.
We created a forward model of dipole-dipole transfer resistances on the
synthetic domain using R2 (http://www.es.lancs.ac.uk/people/amb/Freeware/R2/
R2.htm) to obtain measurement error-free data. Two sets of synthetic data are
generated by adding noise to these data: one with 2% Gaussian noise everywhere, and
the other with 10% Gaussian noise on measurements involving three of the electrodes
on the left (x = 6m, 14m, 22m) and 2% noise everywhere else. The second noisy dataset
was created to simulate the effect of a non-uniform error model that may be typical of
surveys in areas with variable electrode contact or quality.
2.1.2. Boxford Dataset
The first field dataset is from the Boxford Water Meadows Site of Special Scientific
Interest in Berkshire, United Kingdom (Chambers et al., 2014b; Musgrave and Binley,
2011; Uhlemann et al., 2016). The collection of the data was automated using the BGS’s
PRIME system. The ERT array is next to the Northern Array used in Uhlemann et al.
(2016), having 32 electrodes spaced at 0.6 m. A dipole-dipole type measurement
configuration was chosen with dipole lengths (a) of 0.6 m to 2.4 m, and dipole
separation multipliers (n) of 1 to 8. The measurement sequence includes 516 pairs of
reciprocal measurements. Less than 15 minutes was needed to complete the
measurement sequence and each of the measurements is obtained by stacking multiple
readings from the same cycle of current injection to improve signal-to-noise ratio. The
measurement sequence was repeated 96 times within a 24 hour period starting at 5:43
a.m. on 19th November 2015, yielding 96 independent repeats of full reciprocal data.
Approach
128
Each of the repeats has 516 measurements (or pairs of reciprocals). During the 24-hour
period, the air temperature in the area varied between 7 and 10 oC and there was no
recorded precipitation.
2.1.3. Sellafield Dataset
A full-scale 3-D time lapse cross-borehole ERT trial to monitor simulated
subsurface leakage was undertaken at a UK nuclear licensed site in Sellafield, Cumbria,
United Kingdom (Kuras et al., 2016). The data collection setup includes four 40m deep
boreholes and 160 electrodes. The data collection cycle of each ERT frame is less than
a day, and each day’s data includes 51,302 dipole-dipole measurements, including
12,481 pairs of reciprocals. The monitoring spanned a 2-year period with 246 days of
data collection during that time. The first nine months of monitoring includes three
stages of conductive leak simulant injection, while the remainder was designed for
long-term background monitoring. The collection of data was automated using BGS’s
ALERT system. In order to be consistent with the autocorrelation analysis of the
Boxford dataset, we divide the data into two subsets of 96 days (one encompasses all
three injection periods while the other is during long-term background monitoring)
for autocorrelation analysis.
2.2. Analysis methods
2.2.1. Definition of measurement error types
Stacking errors are given by the averaging of ‘stacks’ obtained by the ERT data
collection equipment. Usually they can be output alongside the measured transferred
resistance from the data collection console.
For reciprocal errors, if 𝑅𝑓 is the forward (normal) transfer resistance for a
particular quadrupole and 𝑅𝑟 is the reciprocal of that measurement where its current
and potential dipoles are swapped with the forward measurement, then the mean
absolute transfer resistance (|R|) and absolute errors (|e|) are simply:
|R| =||𝑅𝑓|+|𝑅𝑟||
2 and |e| =
||𝑅𝑓|−|𝑅𝑟||
2. (1)
Approach
129
As a proxy for repeatability errors, the departure from the mean of the j-th repeated
reading for measurement number i (di,j) is given by:
𝑒𝑖,𝑗 = 𝑑𝑖,𝑗 − 𝑑 (2)
where 𝑑 is the mean value for the i-th measurement.
2.2.2. Statistical analysis of measurement errors
The probability density function of an error type for a dataset is obtained by fitting
a Gaussian distribution to the population of errors. Autocorrelation is defined as the
correlation among a sequence of values at a given lag 𝐿:
autocorr(L) =𝐸[(𝑋𝑡−)(𝑋𝑡+𝐿−)]
𝑣𝑎𝑟(𝑋𝑡)=∑ (𝑋𝑡−)(𝑋𝑡+𝐿−)𝑞−𝐿𝑡=1
𝑣𝑎𝑟(𝑋𝑡) (3)
where 𝐸[ ] is the expected value, 𝑋𝑡 is a time-series, 𝑋𝑡+𝐿 is a time-series shifted by lag
𝐿, and and var(𝑋𝑡) are the mean and variance of the time series respectively. q is the
number of repeats for a measurement.
Correlation analysis can be used to study the potential correlation between
measurement errors. The correlation coefficient, r , for the correlation between
arbitrary variables x and y is defined by the products of standard scores (also known
as z-scores or standardized variables) as follows:
𝑟 = 𝑟𝑥𝑦 =1
𝑞−1∑ (
𝑥𝑖−
𝑠𝑥) (
𝑦𝑖−
𝑠𝑦)
𝑞𝑖=1 (4)
For the purposes of our analysis of measurement error correlations, x and y are
series of two measurements that we consider and q is the number of repeats. and
are the means of x and y respectively, while sx and sy are the standard deviations.
2.3. Inversion methods
To obtain 2D tomograms from electrical measurements from the synthetic study
and Boxford site, we use the finite-element based, Occam-type, two-dimensional
electrical resistivity inversion program R2
(www.es.lancs.ac.uk/people/amb/Freeware/R2/R2.htm). The three-dimensional
inversion (Sellafield dataset) was performed by using the commercial code
Approach
130
Res3DInvx64 (Loke and Barker, 1996). The inverse problem is posed as a minimization
problem, where the objective function is defined as
𝛷 = 𝛷𝑑 +𝛷𝑚 = (d − 𝐅(𝐦))𝑇𝐖dT𝐖d(𝐝 − 𝐅(𝐦)) + 𝛼𝐖m
T𝐖m (5)
where d are the data vector (e.g. measured apparent resistivities), F(m) is the set of
simulated data using the forward model and estimated parameters m. Wd is a data
weight matrix, which, if we consider the uncorrelated measurement error case and
ignore forward model errors, is a diagonal matrix with entries equal to the reciprocal
of the standard deviation of each measurement. Forward modelling errors are also
added to the diagonal of Wd. Usually a forward model is run for the computational
grid using a known homogeneous domain. Any discrepancy between the computed
and known apparent resistivity values (i.e. data errors) will be added to the reciprocal
of Wd by means of square root of sum of squares. In this study, we assume
measurements errors are the only source of data errors while other sources, such as
forward modelling errors and field procedural errors, are negligible. To regularize the
minimization problem, a model penalty term 𝛷𝑚 = 𝛼𝐖mT𝐖m is added to impose the
spatial connectedness of the parameter cell values. 𝛼 is a scalar that controls the
emphasis of smoothing.
We can state a desired level of data misfit as 𝛷𝑑 = 𝑁 , where 𝑁 is the number
of measurements (Binley, 2015). In an Occam’s solution, we seek to achieve this desired
data misfit subject to the largest possible value of α. The process is achieved by utilizing
the Gauss–Newton approach, which results in the iterative solution of
(𝐉T𝐖dT𝐖d𝐉 + 𝛼𝐖m
T𝐖m)∆𝐦 = 𝐉T𝐖dT𝐖d(𝐝 − 𝐅(𝐦)) − 𝛼𝐖m
T𝐖m
(6)
𝐦𝑘+1 = 𝐦𝑘 + ∆𝐦
where J is the Jacobian (or sensitivity) matrix, given by 𝐽𝑖,𝑗 = 𝜕𝑑𝑖/𝜕𝑚𝑗 ; 𝐦𝑘 is the
parameter set at iteration k; and ∆𝐦 is the parameter update at iteration k. For the DC
resistivity case, the inverse problem is typically parameterized using log-transformed
resistivities.
Approach
131
The resolution matrix for the inversion is given by:
𝐑 = (𝐉T𝐖dT𝐖d𝐉 + 𝛼𝐖m
T𝐖m)−1𝐉T𝐖d
T𝐖d𝐉 (7)
2.4. Error propagation and uncertainty quantification methods
We follow the Monte Carlo uncertainty propagation procedure of Aster et al.
(2005) outlined below. The goal is to simulate a collection of noisy data vectors and
then examine the statistics of the corresponding models. The advantage of this method
is that it can be readily applied to field data where no repeats are available. The
procedure is achieved by the following steps:
1. Propagate the inverse solution into an assumed noise-free baseline jx1 data
vector 𝐝 (where j is the size of number of measurements) using the forward
model G:
𝐆 = 𝐝 (8)
2. Generate q realizations (i = 1, …, q) of noisy data about using the error model
𝐝i = 𝐝b + 𝛆.∗ 𝐙 (9)
where 𝜺 is the jx1 vector of error levels predicted by the error model and Z is
the standard normal distribution variable and .* is element-wise
multiplication.
3. Invert the q realizations (i = 1, …, q) of noisy data using the inverse model
𝐆𝐦i = 𝐝b + 𝛆i (10)
4. Let A be a q x m matrix where the i-th row contains the departure of the i-th
model from the baseline inverse solution
𝐀i = 𝐦iT − T (11)
5. An empirical estimate of the model covariance matrix is given by
cov() =𝐀T𝐀
𝑞 (12)
6. 95% confidence interval about is given by
Analysis of Errors in Field Datasets
132
± 1.96 ∙ diag(cov())1/2 (13)
7. Similarly, the coefficient of variation of the estimate is given by
diag(cov())1/2./𝑇 (14)
where diag() is a function that extracts the diagonal elements of a matrix and ./ is
element-wise division.
3) Analysis of Errors in Field Datasets
In this section, we report results from the statistical analysis of different types
of errors with the methods outlined in section 2.1 and 2.2. Probability density functions
(PDFs) show the ranges of these errors, while autocorrelation and correlation
coefficient analysis reveals the potential autocorrelation of errors for successive
repeated measurements and correlation of errors between pairs of measurements,
respectively.
3.1. Probability density function of reciprocal and repeatability errors
Before detailed statistical analysis of measurement errors is performed, we first
examine the probability density function of errors obtained from the Boxford dataset.
Since the measurements are repeated 96 times, we can define repeatability errors based
on averaging different numbers of repeats. Figure 2 shows the repeatability errors
based on measurements obtained with a 30 minute, 1 hour, 2 hour, and 24 hour
window. They correspond to averaging 2, 4, 8, and 96 repeats. The mean of the PDF
increases with greater time windows while the variance first decreases, then increases
for the 24 hour repeatability error. When large windows of averaging are used, changes
in the subsurface condition such as diurnal changes in temperature can be mistaken as
errors. This is supported by the observed increase in the mean. For the 24 hour sampled
PDF, the lower tail overlaps that of the 1 hour and 2 hour PDFs while having a much
greater spread. Clearly some measurements do not vary much during the 24 hours
monitoring period while others do: measurements sensitive to the shallower
subsurface will be more susceptible to external influences (e.g. temperature,
evaporation, etc.).
Analysis of Errors in Field Datasets
133
Figure 2 also shows the PDF for stacking errors for each of the measurements
as well as the reciprocal errors from individual datasets. The reciprocal errors PDF
essentially overlay that of the 30 minute repeatability errors. Their similarities may be
explained by the fact that both of them are obtained from averaging pairs of
measurements. It is noteworthy, however, that both the mean and variance of the PDF
for reciprocal errors (which is collected in a 15 minute timeframe) is slightly higher—
which is opposite to our general observation that repeatability errors increase with the
size of the averaging window. Reciprocal errors may be sensitive to other error
contributions not registered by repeatability errors, or the process of taking a reciprocal
measurement introduces an additional source of error.
The stacking errors PDF overlays the low-end of the PDFs of repeatability
errors while having a very small variance. In other words, stacking errors do not
register any of the high-error measurements that appear in the true assessment of
repeatability or in reciprocal errors. For instance, the PDF shows that almost none of
the stacking errors are higher than 10-4 Ω, which covers a majority of the area under
the other PDFs. This shows that stacking errors are potentially an inadequate measure
for describing the true quality of ERT measurements.
The second portion of Figure 2 shows the PDF of stacking, reciprocal, and 2-
week (which correspond to six frames) repeatability errors for the Sellafield dataset. In
general, the ranges of magnitude of the errors are greater due to ground conditions
and contact resistances. Similar to the Boxford results, we find that the stacking errors
are an order of magnitude smaller than reciprocal errors. Since a larger time window
(i.e. days) is used to obtain the repeatability errors, they are significantly greater than
the reciprocal errors.
Analysis of Errors in Field Datasets
134
Figure 2 (a) Comparison of stacking errors, repeatability errors, and reciprocal errors for the Boxford
dataset by plotting probability density functions. The PDFs of reciprocal errors and repeatability errors
are comparable to each other. The stacking errors PDF, however, show very low mean and low variance.
Using stacking errors for measurement errors characterisation may lead to significant underestimation
of uncertainty and over-fitting of data. (b) Comparison of stacking errors, repeatability errors, and
reciprocal errors for the Sellafield dataset. The PDFs for Sellafield show greater variances than those
for Boxford. Since a two-week repeatability cycle is used, the repeatability errors are much greater than
reciprocal errors. In general, the stacking errors are more than an order-of-magnitude smaller than the
reciprocal errors, indicating there may be significant underestimation of errors if they are used as error
weights. The mean and standard deviation of each fitted normal distribution is shown next to the
legend.
3.2. Autocorrelation analysis
Autocorrelation analysis is used to investigate whether there is “memory” (i.e.
correlation in time) in ERT measurement errors. We compare autocorrelation plots
between the (i) departure from the mean and (ii) reciprocal errors of individual
measurements for the Boxford dataset in Figure 3. Each grey translucent line
represents the autocorrelation function of a measurement, while the red line is the
mean averaged across all measurements. The red hashed regions highlights the area
with an autocorrelation value below the critical Pearson’s correlation coefficient
(Pearson and Hartley, 1970), which is around 0.2 for 96 timesteps. For the departure
Analysis of Errors in Field Datasets
135
from the mean, the autocorrelation drops to 0.5 on average at lag 1 and then decreases
steadily. This is likely to be due to the presence of diurnal temperature effects within
the 24 hour data collection cycle. Individual reciprocal errors, show negligible
autocorrelation for all lag numbers (i.e. within the hashed region). Thus, we can
conclude the individual reciprocal errors between any two repeated measurements are
independent from one another for this survey. From the above, we see that the
assumption of uncorrelated errors is appropriate for reciprocal errors but not so much
for long-term repeatability errors.
Figure 4 shows the autocorrelation of (a) departure from the mean and (b)
reciprocal errors for the 96 datasets collected continuously at the Sellafield site
encompassing the three injection periods (22th Jan 2013 – 3rd Nov 2013) and those for
another 96 datasets during the long-term background monitoring period (i.e. no
injection, 5th Nov 2013 – 31st Mar 2014). We can see much greater autocorrelation of
errors at Sellafield than at Boxford. Like in the Boxford dataset, the departure from the
mean shows greater autocorrelation than individual reciprocal errors, both for
injection and long-term background monitoring. In general, however, the departure
from the mean and reciprocal errors during background monitoring reach insignificant
autocorrelation sooner than during injections. While the 96 datasets at Boxford were
collected in less than 24 hours, the two groups of 96 datasets examined above were
collected over a period of months. It is certain that the subsurface condition had
changed during the monitoring period due to injection, dilution and dispersal of tracer,
as well as regional groundwater and vadose zone changes (see Kuras et al. (2016) for
details).
Analysis of Errors in Field Datasets
136
Figure 3 Autocorrelation of (a) departure from the mean (as a measure of repeatability errors) and (b)
reciprocal errors for the 96 datasets collected continuously continuously within 24h at the Boxford site.
The number of lags is on the horizontal axis (here 1 lag = 15 minutes). Each grey translucent line plots
the autocorrelation of one of the 516 ERT measurements as a function of lag. The red line denotes the
mean autocorrelation. For each autocorrelation plot, 96 datasets are considered. The hashed region has
insignificant correlation according to the critical Pearson’s test (around ±0.2).
Figure 4 Autocorrelation of (a) departure from the mean (as a measure of repeatability errors) and (b)
reciprocal errors for the 96 datasets collected continuously at the Sellafield site encompassing the three
injection periods (22/1/2013 – 3/11/2013). The number of lags is on the horizontal axis (here 1 lag = ~2 to
3 days). Each grey translucent line plots the autocorrelation of one of the 12481 ERT measurements as
a function of lag. The red line denotes the mean autocorrelation. For each autocorrelation plot, 96
datasets are considered. The hashed region has insignificant correlation according to the critical
Pearson’s test (around ±0.2). Similarly, (c) and (d) show the same for the long-term background
monitoring period (i.e. no injection, 5/11/2013 – 31/3/2014).
A New Error Model
137
3.3. Correlation coefficient analysis
Although measurement errors are commonly assumed to be uncorrelated in
ERT, previous studies have highlighted the potential of correlation in measurement
errors because ERT surveys typically use the same electrodes for multiple
measurements (Ramirez et al., 2005). We have computed the correlation coefficient
matrix for departure from the mean and reciprocal errors for the Boxford dataset. We
subdivide all the correlation coefficients into two groups: one group consists of pairs
of measurements that share one or more electrodes and the other consists of all
measurement pairs. Next, we grouped departure from the mean or reciprocal errors as
a function of dipole-dipole separation multiplier n and plot the mean of each group.
We show in Figure 5 that for all n used for the Boxford dataset, the mean correlation
coefficients for measurement pairs that share one or more electrodes are always higher
than the means for all pairs. The mean correlation coefficients for reciprocal errors are
orders of magnitude smaller than those of departure from the mean. The effect of
electrode sharing is also pronounced for reciprocal errors—the mean correlations of all
reciprocal errors pairs are negligible while those for pairs that share one or more
electrodes are consistently higher. Note that electrode sharing only occurs in ~10% of
all pairs. Figure 5 shows that by taking into account the correlation of the electrodes
used to make multiple measurements, ERT measurement errors may be better
modelled. With the autocorrelation results, we also show that the departure from the
mean exhibits more spatial and temporal correlation than the reciprocal errors.
4) A New Error Model
4.1. Model definition and implementation
Our error analysis reported in section 3) revealed that the combination of
electrodes used appears to influence ERT measurement errors. Therefore, we
developed a new error model based on linear mixed effects (LME) models to group
measurement errors by the electrodes used to obtain them, which allows us to
incorporate the effects of electrode combinations.
A New Error Model
138
Figure 5 Mean correlation coefficient of departure from the mean (as a measure of measurement errors)
and reciprocal errors for measurement pairs from the Boxford dataset as a function of dipole separation
multiplier n. For both departure from the mean and reciprocal errors, mean correlation coefficients are
distinctively higher for measurements that share electrode(s) in their quadrupoles than the mean
correlation coefficients for all measurements, indicating by considering the effect of using each
electrode to make multiple measurements may improve error models. Also, note that the reciprocal
errors have strikingly lower correlation coefficients than the departure from the mean. Note that
electrode sharing only occurs in ~10% of all pairs.
The linear mixed effect model is a powerful statistical tool in settings where
repeated measurements are made on the same statistical units (longitudinal study), or
where measurements are made on clusters of related statistical units (Bates et al., 2015;
Diggle et al., 2015; Pinheiro and Bates, 1988; West et al., 2007). It is especially useful to
group qualitative variables that influence the data. In general, a mixed effect model is
given by
A New Error Model
139
𝐲 = 𝐗𝛃⏟𝑓𝑖𝑥𝑒𝑑
+ 𝐙𝐛⏟𝑟𝑎𝑛𝑑𝑜𝑚
+ 𝛆⏟𝑒𝑟𝑟𝑜𝑟
(15)
where y is the n-by-1 response vector (i.e. dependent variable), and n is the number of
observations; X is an n-by-p fixed-effects design matrix, and p is the number of fixed
effect variables; β is a p-by-1 fixed-effects vector and q is the number of random effect
variables; Z is an n-by-q random-effects design matrix; and 𝐛 is a q-by-1 random-effects
vector. ε is an n-by-1 unknown vector of random, independent and identically
distributed errors. In broad terms, fixed-effects are variables that are expected to have
an effect on the dependent variable (i.e. explanatory variables in linear regression),
while random effects are categorical grouping factors.
Linear mixed effect models can now be readily fitted using the MATLAB®
statistics and machine learning toolbox and the lme4 package for R (Bates et al., 2015).
In this paper, we model measurement errors in ERT by treating transfer resistances as
fixed effects and each of the electrodes used (A, B, M, N) as grouping variables. The
above model was implemented in MATLAB® (see supplementary information for
more details).
The linear mixed effect model essentially establishes a hierarchy or grouping
when fitting the measurement errors. Fitting is achieved by both optimizing fit within
each cluster, while the covariate vectors link the fixed and random effects between
clusters. The clustering introduces additional degrees of freedom that allow a better fit
of measurement errors than commonly used linear models. An illustrative example of
the LME grouping formulation can be found in the supplementary information, along
with details for fitting the LME error model to the Boxford and Sellafield field datasets.
The evolution of the error model coefficients with time is also described.
4.2. LME error model behaviour for time-lapse ERT measurements
A longitudinal survey is a correlational research study that involves repeated
observations of the same variables over long periods of time. One of the original uses
of LME models is to handle longitudinal data in tracking studies to eliminate potential
bias of using the same samples. For example, in a drug study the health indicators of
the same group of patients are sampled multiple times during a long period. The times
Comparison of Error Models using Image Appraisal
140
at which they are sampled can be used as an additional grouping variable in the LME
model. With the increased popularity of long-term monitoring using ERT and other
geophysical methods, it may be beneficial to treat measurement errors as longitudinal
data too. In Figure 6, we compare fitting observed measurement errors in the 96 repeat
datasets from Boxford individually (i.e. obtaining 96 LME equations) and as
longitudinal data (i.e. obtaining one LME equation, with the repeat number as an
additional grouping variable). The scatter plots show that a much better fit is obtained
by fitting each of the 96 datasets individually. In other words, treating ERT
measurement errors as longitudinal data does not better characterise them.
Measurement errors should instead be characterised on a frame-by-frame basis for
long-term geophysical monitoring. Of course, this comparison does not involve the
same degree of freedoms for both methods, which should be repeated using more
robust criteria (e.g. Bayesian information criteria).
Figure 6 Comparison of fitting reciprocal errors of time-lapse data as (a) individual datasets, fitting
each dataset individually with a different LME model and (b) longitudinal data, fitting all data with
one LME model. The above shows that it is much better not to treat errors as longitudinal data.
5) Comparison of Error Models using Image Appraisal
Improvements in the measurement error model are only useful if they can lead
to better inversion results. We applied the new error model to the synthetic data and
field data from the Boxford and Sellafield sites. Also, we will consider the resolution
Comparison of Error Models using Image Appraisal
141
matrix and model variance from Monte Carlo simulations to see whether using the
new error model can give additional insight to data and reduce uncertainty.
5.1. Synthetic data
Using the synthetic domain introduced in section 2.1, we compared the
inversion results and the corresponding resolution matrices and uncertainty estimates
using different error models. Note that since Figure 1 and Figure 7 use the same
domain and have the same resistivity structure, Figure 1(c) can be seen as a benchmark
case where the data is inverted with perfect knowledge of measurement errors.
Figure 7 (a – c) shows inversion results for synthetic data where measurements
involving three “bad electrodes” are corrupted by 10% noise and others by 2% noise.
We first compare the inversion with two linear error model—one assumes there are no
bad electrodes (i.e. the 2% error model), while the other is obtained by fitting the
corrupted data with the Koestel et al. (2008) model (i.e. the 4.52% linear model). We
see that the resultant resistivity model from assuming the 2% linear error model is very
noisy while that from assuming 4.52% linear error model is smoother. With the LME
error model, however, the inversion result is the most similar to that of the benchmark
case (Figure 1c) (see also rms errors printed on plots). The effect of better
characterisation of measurement errors by the LME model is manifested in the
inversion results.
Figure 8 (a – c) shows the diagonal terms of the resolution matrices for the
inversion using (a) 2% linear, (b) 4.52% linear, and (c) LME error models. In general,
the resolution patterns are uniform laterally yet decreases with depth. For the 2% linear
error model, we see that some of the artefacts from the inversion results are also shown
on the resolution pattern. For the LME error model, the resolution on the right is
somewhat higher than on the left for the top layer, where the bad electrodes are located.
The resolution values are between that of (a) and (b) in most of the cells, although some
of the cells near the surface show very high resolution. The above shows that while the
resolution from the linear error model is purely a function of distance away from
sources and sensors and therefore cannot distinguish quality between measurements,
Comparison of Error Models using Image Appraisal
142
the LME error model allows the inversion to resolve areas unaffected by the bad
electrodes better.
Subsequently, we ran Monte Carlo experiments using the procedure in section
2.4 to understand how uncertainty in measurement errors propagates to affect
uncertainty in the parameter estimates. The Monte Carlo experiment results can be
used to form empirical model covariance matrices. This matrix shows how information
is shared between parameters (i.e. model estimate of different elements). In the ideal,
noise-free and well-defined case, the model covariance matrix should be a zero matrix,
meaning the parameter is deterministically known and the parameters are not
correlated with one another. Figure 9 (a – b) show that assuming a 2% linear error
model yields lower model covariances than the 4.52% model, which is expected
because lower percentage error implies less sharing of data. Also, the band of high
covariances is also narrower. With the LME model (Figure 9c), however, we notice that
the model covariances values are lower than those of the 2% and 4.52% models. More
importantly, the spread of the high covariance region is less uniform than the linear
models, meaning that only measurements affected by the bad electrodes share
information heavily with others. The above agrees with the comparison of resolution
matrices—the new error model is able to exploit information in noisy data without
increasing the overall noise level.
The diagonal term of the empirical model covariance matrix (i.e. variance)
shows the variability among parameter estimates from multiple Monte Carlo
simulation realizations. Specifically, the higher a diagonal term, the more uncertain is
the estimate. We plot their ranges in Figure 8 (d – f). For all three error models, the
variability is always the smallest at depth because deeper regions are less well resolved
for surface arrays. As a result, the model estimates at greater depths are closer to the
initial guess values and therefore, there is less difference between the realizations of
Monte Carlo model estimates. In Figure 8 (g – i), we plot the model-averaged
parameter estimates. The transparency of each element is inversely proportional to its
model variance, as shown in Figure 8 (d – f). In other words, the elements that have
more variable or uncertain estimates have greater transparency. The inversion results
Comparison of Error Models using Image Appraisal
143
from assuming a 2% error model are more variable than for the 4.52% model. Model
averaging also smooths out the noisy artefacts from deterministic inversion (compare
Figure 8g and Figure 7a). The LME error model gives the most reliable model estimates
among the three error models tested. Also, it is worth noting that the model-averaged
parameter estimates are comparable to that obtained from deterministic inversion.
This means that with the LME error model, there is no need to run many realizations
of the inverse model in order to obtain reliable parameter estimates. Importantly,
inversion using the LME error model gives the highest resolution and the least model
variance (Figure 8), and reduces uncertainty in inversion results.
Figure 7 Synthetic surface ERT experiments to demonstrate the performance of the error models. For
data involving 3 bad electrodes(marked by “X”), data is corrupted by 10% white noise while for the
rest of the data 2% white noise is added. (a) Inverted resistivity distribution using the 2% linear error
model (b) Inverted resistivity distribution using a 4.52% (obtained from the Koestel et al. (2008) method)
linear error model (c) Inverted resistivity distribution using the LME error model. Note that the
convergence target for all the inversions is a chi-squared statistic of 1.
Comparison of Error Models using Image Appraisal
144
Figure 8 (a – c) Diagonal of resolution matrix for inversion using the following error models for inverting the synthetic data corrupted by “bad electrodes”: (a) 2% linear
model (b) 4.52% linear model (c) LME model. (d - f) variance of element-wise log-resistivity estimates using each of the error models obtained from Monte Carlo experiments.
The colour scale is the same for all three error models. Darker cells indicate more similar model estimates between Monte Carlo estimates. (g - i) mean model estimates
from Monte Carlo experiments. The transparency is controlled linearly by the variance shown in (d – f). With model averaging, the mean estimates of the three error models
are similar. It is noted, however, the deterministic results from the LME model agrees the best with its model-averaged results.
Comparison of Error Models using Image Appraisal
145
Figure 9 Empirical model covariance matrix using the Monte Carlo uncertainty propagation procedure
and the following error models: (a) 2% linear model (b) 4.52% linear model (c) LME model. The size of
the matrix is m x m, where m is the number of model parameters. By comparing (a) and (b), it is shown
that assuming higher error levels, there is higher covariance between model parameters. With the LME
error model, the model covariance is the lowest. While the spread of high covariance entries are quite
even throughout the matrix, we can see that the spread for (c) is quite uneven: generally, elements on
the left of the domain have higher spread.
5.2. Boxford field data
In Figure 10, we compare the inversion results of field data for the Boxford
datasets. When using reciprocal data, we only consider one of the 96 available datasets
(i.e. the first of the 96 repeats). The resultant image from using linear or LME error
models for reciprocal or 24-h repeatability errors (not shown) for the Boxford dataset
are effectively identical. When the linear model is applied to the stacking errors, the
resultant image shows a rather sharp feature. Surprisingly, when the LME model is
applied to the stacking errors, there is no distinguishable difference between its result
and those using reciprocal or 24-h repeatability errors. This shows that although we
have shown above and warned against the potential underestimation of measurement
errors caused by using the stacking errors, the LME error model is capable of
minimizing such effects. We suspect that because of the low mean and low variance of
the stacking errors, the linear error model is forced to assign very low errors across the
dataset. The LME error model, in contrast, has more degrees of freedom to better fit
the observed stacking errors. Note that this does not guarantee physically more
realistic results.
This finding has significant implications because all modern ERT equipments
output stacking errors and these do not require additional data collection time. For
many existing datasets where only stacking errors are available or in applications
Comparison of Error Models using Image Appraisal
146
where the collection of repeats and reciprocal is prohibitive, we recommend using a
LME error model instead of a linear model for the stacking errors.
Figure 10 Inversion results from Boxford using (a) linear error model for stacking errors, (b) LME error
model for stacking errors, (c) linear error model for reciprocal errors.
5.3. Sellafield field data
We inverted the Sellafield data collected on 5th February 2013, which was two
days before the first tracer injection (Kuras et al., 2016). Of the 51,302 measurements in
the sequence, there are 12,412 pairs of valid reciprocal measurements. We fitted them
with the LME error model. Note that we have not removed any high-error outliers.
Figure 11 shows the resultant 3-D static inversion image and its associated uncertainty
estimates (model standard deviation and model coefficient of variation) derived from
Monte Carlo simulations. The resultant model clearly delineates zones of high and low
resistivities. In terms of uncertainty, regions next to the borehole and towards the top
of the monitoring array have significantly higher model standard deviation.
Compared with the absolute images of resistivity reported in Kuras et al. (2016) (note
that we use the same mesh and inversion code), Figure 11a shows similar patterns but
with a smaller range and variations in resistivities.
Discussion
147
Figure 11 (a) 3-D static deterministic inversion results from Sellafield on 5th February, 2013. Error
weights are prescribed by fitting an LME error model. Black lines are boreholes installed with
electrodes. (b) The corresponding uncertainty estimates obtained from Monte Carlo simulations, given
by model standard deviation from Monte Carlo experiments. (c) The corresponding coefficient of
variation of Monte Carlo model estimates.
6) Discussion
In the present study, we have used statistical methods to explore ways to
improve the current practice of modelling measurement errors in ERT. Among them,
we have found that the correlation coefficients of measurement pairs that share some
of the electrodes are consistently higher than average. Therefore, we have developed a
Discussion
148
new error model that considers such effects in ERT surveys by adding electrode-
specific fitting terms (i.e. the LME error model).
The proposed error model based on the linear mixed effect (LME) model shows
superior performance in terms of characterising errors when compared against an
unknown linear error model. The LME model assumes that errors are linearly
dependent on transfer resistances and employs the electrodes used to make each
measurement as grouping variables. The LME error model can more accurately predict
observed measurement errors. However, as we have already argued in section 4.2,
individual errors should not be used directly for inversion because in most practical
situations they are only averages between two points. To improve the robustness of
the linear error models, errors can be grouped by the magnitude of transfer resistances
(Koestel et al., 2008). Such binning, however, is arbitrary and the resultant error models
can be sensitive to the number of bins used. The LME error model is based on the same
idea of grouping, yet it considers all of the four electrodes that are used to make each
dipole-dipole quadrupole measurement and uses them as the grouping variable.
Electrode number is a qualitative variable and it is a reasonable assumption that each
electrode has slightly different quality.
The patterns of resolution matrix and model covariance matrix associated with
using the LME error model are different from those using the linear model. This has
important implications for inversion and uncertainty estimation because it shows that
the LME model is capable of detecting poorer measurements and downweighting
them in an inversion. Most inversion schemes are capable of weighting data according
to their quality. Yet in common ERT practice, either uniform percentage errors (i.e. a
linear model) are assumed or the errors are not characterised at all. The LME error
model is one of the first statistical tools to characterise the variable quality of ERT
measurements (while not using individual errors directly) so that the data weighting
schemes in inverse models can be fully utilized.
While fitting a LME model for each set of reciprocal errors gives promising
estimates, fitting time series of reciprocal errors with a single LME model and using
the sequence of data collection as an additional grouping variable (i.e. as longitudinal
Discussion
149
data) can yield inferior results. Evaluation of the individual resultant LME error
models reveals that, for the dataset considered here, the fixed and random effects
coefficients vary over the 24-hour period. Such results challenge our common
assumption that electrode quality is extremely stable. The laboratory study by
LaBrecque and Daily (2008) on the measurement errors of 15 electrode materials
showed many possibilities for electrode quality to evolve during the course of a ERT
experiment, some even in the timescale of minutes. Therefore, taking many repeats for
measurements probably will not provide better error estimates because electrode
quality may evolve during the process. In summary, we recommend the collection of
reciprocal measurements at each timeframe and fit a LME model based on the
measured transfer resistance and electrodes used to capture the minor variations in
electrode quality during ERT experiments.
We have found in the Boxford inversion results that there is no distinguishable
difference between using repeatability and reciprocal errors in inversions (figures not
shown). From the PDFs, the stacking errors are much smaller and much less variable
than the repeatability or reciprocal errors at Boxford. With the linear error model, the
resultant image for using stacking errors is noisy. With the LME error model, however,
the inversion image is comparable to that obtained from using repeatability or
reciprocal errors. We attributed its better results to the better handling of spurious and
overly optimistic estimates of errors by the LME error model.
For the Sellafield dataset, we demonstrated the application of the new LME
error model to model reciprocal errors and used its predicted errors for 3-D inversion
and uncertainty quantification (i.e. model variance). Such uncertainty estimates are
useful as they visualize how uncertainties in measurements propagate to uncertainties
in the inverse model estimate.
We have highlighted in the previous section that the new LME error model can
be widely applied to essentially any ERT inversion algorithms. It better predicts errors
that are used to prescribe weights of the data weighting or covariance matrix. The
resultant matrix remains diagonal so that it does not increase computation costs during
inversion. Unlike the data quality control strategies recently proposed by Deceuster et
Discussion
150
al. (2013) and Mitchell and Oldenburg (2016), the new LME error model can be applied
to any static and time-lapse ERT problems regardless of their size. Since the model
considers the effect of the variable quality of electrodes, it requires minimum culling
of data or re-inversion. Alternatively, the new LME error model can be used alongside
with other data quality control strategies.
The flexibility of the LME model allows it to be applied to characterisation of
errors in other geophysical measurements. For example, geophones used in seismic
tomography can be used as grouping variables for their errors. A straightforward next
step for future study would be to extend the LME error model to induced polarisation
(IP) studies. It has been reported in the literature that IP surveys are even more
sensitive to electrode configuration than ERT. Much recent work has been done to
improve quality of IP measurements. For example, Dahlin et al. (2013) conducted a
duplicate IP survey for a planned tunnel using two types of cable spreads: one with
standard multi-core cables and the other with separate cable spreads for transmitting
current and measuring potentials. They suggest that the single cable spread is
sufficient to give good IP data but suggest the use of separate cable spread for spectral
IP inversion and recovery of Cole-Cole parameters. Flores-Orozco et al. (2012)
quantified the measurement errors in spectral IP imaging and established a new phase
error model. It is an extension of previous models where the discrepancy between
normal and reciprocal measurements is analysed (Binley et al., 1995; LaBrecque et al.,
1996a; Slater and Binley, 2006). They also conducted a bin analysis to ensure the
assumption of a normal distribution of errors is valid and showed that, for spectral IP
measurements, phase error discrepancies show a consistent behaviour for all
frequencies. They proposed an inverse power-law relationship between the error on
phase and the corresponding resistance. This brief review highlights the similarities
between ERT and IP measurement error models and we believe that the proposed LME
model can improve IP measurement errors characterisation, too. Future studies should
consider applying the LME error model.
Finally, the proposed LME models can be used readily in Bayesian
formulations for ERT inversion. The LME error model can be used to prescribe entries
Conclusion and recommendations
151
of the data covariance matrix in their likelihood functions, which are usually assumed
to be diagonal for computational convenience. Note that the LME method considers
errors due to electrodes used as a grouping variable rather than enforcing a correlation
function, which would lead to a full data covariance that is computationally difficult
to invert. By treating the potential correlation of electrode effects as grouping variables
instead, the data weighting or covariance matrix remains diagonal; furthermore, strict
and unnecessary assumptions on the correlation between measurements are avoided.
7) Conclusion and recommendations
Our analysis of field datasets shows that short-term repeatability and reciprocal
errors are very comparable, while stacking errors are significantly lower. Repeatability
errors, however, may increase over time because of subsurface changes between
repeats. Repeatability errors also tend to show greater autocorrelation in time for the
same measurements, as well as correlation between measurements, than reciprocal
errors. Stacking errors are found to have significantly lower magnitude and variability,
indicating it may be an overly optimistic measure of measurement error. Correlation
coefficients between pairs of measurements that share some of the electrodes used are
higher than pairs that use completely different electrodes. This confirms speculations
from previous studies that the common use of electrodes may contribute to some
correlation in errors (Ramirez et al., 2005).
Based on our error analysis, we confirm the value of collecting reciprocal data
in ERT studies, although when making reciprocal measurements, care should be taken
to avoid electrode charge-up effects (Dahlin, 2000; Wilkinson et al., 2012). If it is too
difficult to set up reciprocal measurements, we recommend running a duplicate survey
immediately after the completion of the original survey. Long-term repeatability data
does not bring extra benefits for fitting error models because subsurface conditions
may change over time. But they may be very useful for long-term quality assurance,
for example, detecting instrument drift or abnormal system behaviour. Stacking errors
should be avoided when assigning error weights because of their low magnitude and
low variability. For modelling the measurement errors, we recommend fitting a linear
Acknowledgements
152
mixed effect (LME) model over the commonly used linear model. The new LME error
model uses both the combination of electrodes used for making ERT measurements
and the proportional relationship between errors and transferred resistance in order to
better characterize measurement errors. Our synthetic example shows that the LME
error model is capable of picking up errors due to the varying quality of electrodes and
adjusts resolutions in the inverse model accordingly. This is different from the
traditional linear model approach where the resolution everywhere in the entire
inverse model domain has to reduce. The new LME model not only improves the
inversion results, but also reduces the uncertainty (i.e. variance) in the model estimates.
For time-lapse data, we recommend fitting a LME model for each time step because its
coefficients change over time and fitting all the data from the different time steps with
a single LME model (i.e. as longitudinal data) yields inferior results. We have
demonstrated the applicability of the above-recommended procedure by fitting the
LME model to errors observed in two field datasets and inverting the data. This
procedure is easy to implement and requires minimal changes to the current practice.
Widely implementing this procedure in future geophysical studies can greatly
improve their overall reliability—a necessary step towards obtaining more
quantitative information from geophysical methods across a range of disciplines and
applications.
8) Acknowledgements
This work is supported by a Nuclear Decommissioning Authority PhD Bursary
awarded to the first author and an accompanying PhD studentship provided by
Lancaster University. This paper is published with the permission of the Nuclear
Decommissioning Authority, Sellafield Ltd., and the Executive Director of the British
Geological Survey (NERC). We thank Paul Mclachlan (Lancaster University) for
assistance on using R. We are grateful to two anonymous reviewers for the
constructive reviews of the manuscript.
References
153
9) References
Aster, R., Borchers, B., Thurber, C.H., 2005. Parameter estimation and inverse problems.
Elsevier, Burlington, Massachusetts, USA.
Barker, J.A., 1991. The reciprocity principle and an analytical solution for Darcian flow
in a network. Water Resour. Res. 27, 743–746. https://doi.org/10.1029/91WR00258
Bates, D., Maechler, M., Bolker, B.M., Walker, S., 2015. Fitting Linear Mixed-Effects
Models using lme4. J. Stat. Softw. 67, 1–48. https://doi.org/10.18637/jss.v067.i01
Beven, K., Westerberg, I., 2011. On red herrings and real herrings: disinformation and
information in hydrological inference. Hydrol. Process. 25, 1676–1680.
https://doi.org/10.1002/hyp.7963
Binley, A., 2015. Tools and techniques: Electrical methods. Treatise Geophys.
https://doi.org/10.1016/B978-0-444-53802-4.00192-5
Binley, A., Hubbard, S.S., Huisman, J.A., Revil, A., Robinson, D.A., Singha, K., Slater,
L.D., 2015. The emergence of hydrogeophsyics for improved understanding of
subsurface processes over multiple scales. Water Resour. Res. 51, 3837–3866.
https://doi.org/10.1002/2015WR017016
Binley, A., Ramirez, A.L., Daily, W., 1995. Regularised image reconstruction of noisy
electrical resistance data, in: Proceedings of the 4th Workshop of the European
Concerted Action on Process Tomography,. Bergen, pp. 401–410.
Bruggeman, G.A., 1972. The reciprocity principle in flow through heterogeneous
porous media, in: Fundamentals of Transport Phenomena in Porous Media,.
Elsevier, New York, p. 136.
Cassiani, G., Bruno, V., Villa, A., Fusi, N., Binley, A.M., 2006. A saline tracer test
monitored via time-lapse surface electrical resistivity tomography. J. Appl.
Geophys. 59, 244–259. https://doi.org/10.1016/j.jappgeo.2005.10.007
Chambers, J.E., Wilkinson, P.B., Uhlemann, S., Sorensen, J.P.R., Roberts, C., Newell,
A.J., Ward, W.O.C., Binley, A., Williams, P.J., Gooddy, D.C., Old, G., Bai, L., 2014.
Derivation of lowland riparian wetland deposit architecture using geophysical
image analysis and interface detection. Water Resour. Res. 50, 5886–5905.
https://doi.org/10.1002/2014WR015643
Crook, N., Binley, A., Knight, R., Robinson, D.A., Zarnetske, J., Haggerty, R., 2010.
Electrical resistivity imaging of the architecture of substream sediments. Water
Resour. Res. 46, W00D13. https://doi.org/10.1029/2008WR006968
References
154
Dahlin, T., 2000. Short note on electrode charge-up effects in DC resistivity data
acquistiion using multi-electrode arrays. Geophys. Prospect. 48, 181–187.
https://doi.org/10.1046/j.1365-2478.2000.00172.x
Dahlin, T., Dalsegg, E., Sandstrom, T., 2013. Data quality quantification for time
domain IP data acquired along a planned tunnel near Oslo, Norway, in: Near
Surface Geoscience 2013--19th European Meeting of Environmental and
Engineering Geophysics. Bochum, Germany.
Day-Lewis, F.D., Johnson, C.D., Singha, K., Lane Jr, J.W., 2008. Best practices in
electrical resistivity imaging: Data collection and processing, and application to
data from Corinna, Maine. EPA report, Boston, MA.
Deceuster, J., Kaufmann, O., Camp, M. Van, 2013. Automated identification of changes
in electrode contact properties for long-term permanent ERT monitoring
experiments. Geophysics 78, E79–E94. https://doi.org/10.1190/GEO2012-0088.1
Delay, F., Ackerer, P., Guadagnini, A., 2011. Theoretical analysis and field evidence of
reciprocity gaps during interference pumping tests. Adv. Water Resour. 34, 592–
606. https://doi.org/10.1016/j.advwatres.2011.02.006
Diggle, P.J., Heagerty, P., Liang, K.-Y., Zeger, S., 2015. Analysis of Longitudinal Data,
2nd ed. Oxford University Press.
Falade, G.K., 1981. Analysis of the reciprocity concept in a porous medium. Water
Resour. Res. 17, 918–920. https://doi.org/10.1029/WR017i004p00918
Flores-Orozco, A., Kemna, A., Zimmermann, E., 2012. Data error quantification in
spectral induced polarization imaging. Geophysics 77, E227.
https://doi.org/10.1190/geo2010-0194.1
Flores-Orozco, A., Williams, K.H., Long, P.E., Hubbard, S.S., Kemna, A., 2011. Using
complex resistivity imaging to infer biogeochemical processes associated with
bioremediation of an uranium-contaminated aquifer. J. Geophys. Res.
Biogeosciences 116, 1–17. https://doi.org/10.1029/2010JG001591
French, H.K., Hardbattle, C., Binley, A., Winship, P., Jakobsen, L., 2002. Monitoring
snowmelt induced unsaturated flow and transport using electrical resistivity
tomography. J. Hydrol. 267, 273–284. https://doi.org/10.1016/S0022-
1694(02)00156-7
Gélis, C., Revil, A., Cushing, M.E., Jougnot, D., Lemeille, F., Cabrera, J., de Hoyos, A.,
Rocher, M., 2010. Potential of electrical resistivity tomography to detect fault
zones in limestone and argillaceous formations in the experimental platform of
References
155
Tournemire, France. Pure Appl. Geophys. 167, 1405–1418.
https://doi.org/10.1007/s00024-010-0097-x
Haarder, E.B., Jensen, K.H., Binley, A.M., Nielsen, L., Uglebjerg, T.B., Looms, M.C.,
2015. Estimation of recharge from long-term monitoring of saline tracer transport
using electrical resistivity tomography. Vadose Zo. J.
https://doi.org/10.2136/vzj2014.08.0110
Hayley, K., Bentley, L.R., Gharibi, M., 2009. Time-lapse electrical resistivity monitoring
of salt-affected soil and groundwater. Water Resour. Res. 45, 1–14.
https://doi.org/10.1029/2008WR007616
Henderson, R.D., Day-Lewis, F.D., Abarca, E., Harvey, C.F., Karam, H.N., Liu, L., Lane,
J.W.J., 2010. Marine electrical resistivity imaging of submarine groundwater
discharge: sensitivity analysis and application in Waquoit Bay, Massachusetts,
USA. Hydrogeol. J. 18, 173–185. https://doi.org/10.1007/s10040-009-0498-z
Hermans, T., Vandenbohede, A., Lebbe, L., Nguyen, F., 2012. A shallow geothermal
experiment in a sandy aquifer monitored using electric resistivity tomography.
Geophysics 77, B11. https://doi.org/10.1190/geo2011-0199.1
Huisman, J.A., Rings, J., Vrugt, J.A., Sorg, J., Vereecken, H., 2010. Hydraulic properties
of a model dike from coupled Bayesian and multi-criteria hydrogeophysical
inversion. J. Hydrol. 380, 62–73. https://doi.org/10.1016/j.jhydrol.2009.10.023
Irving, J., Singha, K., 2010. Stochastic inversion of tracer test and electrical geophysical
data to estimate hydraulic conductivities. Water Resour. Res. 46, W11514.
https://doi.org/10.1029/2009WR008340
Johnson, T.C., Slater, L.D., Ntarlagiannis, D., Day-Lewis, F.D., Elwaseif, M., 2012.
Monitoring groundwater-surface water interaction using time-series and time-
frequency analysis of transient three-dimensional electrical resistivity changes.
Water Resour. Res. 48, 1–13. https://doi.org/10.1029/2012WR011893
Johnson, T.C., Versteeg, R., Thomle, J., Hammond, G., Chen, X., Zachara, J., 2015. Four-
dimensional electrical conductivity monitoring of stage-driven river water
intrusion: Accounting for water table effects using a transient mesh boundary
and conditional inversion constraints. Water Resour. Res. 51, 6177–6196.
https://doi.org/10.1002/2014WR016129
Kemna, A., 2000. Tomographic inversion of complex resistivity. Ruhr-Universität
Bochum.
References
156
Kim, J., Supper, R., Ottowitz, D., Jochum, B., Yi, M., 2016. A new measurement protocol
of direct current resistivity data. Geophysics 81.
Koestel, J., Kemna, A., Javaux, M., Binley, A., Vereecken, H., 2008. Quantitative
imaging of solute transport in an unsaturated and undisturbed soil monolith
with 3-D ERT and TDR. Water Resour. Res. 44, 1–17.
https://doi.org/10.1029/2007WR006755
Kuras, O., Wilkinson, P.B., Meldrum, P.I., Oxby, L.S., Uhlemann, S., Chambers, J.E.,
Binley, A., Graham, J., Smith, N.T., Atherton, N., 2016. Geoelectrical monitoring
of simulated subsurface leakage to support high-hazard nuclear
decommissioning at the Sellafield Site, UK. Sci. Total Environ. 566–567, 350–359.
https://doi.org/10.1016/j.scitotenv.2016.04.212
LaBrecque, D.J., Mletto, M., Daily, W., Ramirez, A.L., Owen, E., 1996. The effects of
noise on Occam’s inversion of resistivity tomography data. Geophysics 61, 538.
https://doi.org/10.1190/1.1443980
Lesparre, N., Boyle, A., Grychtol, B., Cabrera, J., Marteau, J., Adler, A., 2016. Electrical
resistivity imaging in transmission between surface and underground tunnel for
fault characterization. J. Appl. Geophys.
https://doi.org/10.1016/j.jappgeo.2016.03.004
Linde, N., Renard, P., Mukerji, T., Caers, J., 2015. Geological Realism in
Hydrogeological and Geophysical Inverse Modeling: a Review. Adv. Water
Resour. 86, 86–101. https://doi.org/10.1016/j.advwatres.2015.09.019
Loke, M.H., Barker, R.D., 1996. Practical techniques for 3D resistivity surveys and data
inversion. Geophys. Prospect. 44, 499–523. https://doi.org/10.1111/j.1365-
2478.1996.tb00162.x
Meyerhoff, S.B., Maxwell, R.M., Revil, A., Martin, J.B., Karaoulis, M., Graham, W.D.,
2014. Characterization of groundwater and surface water mixing in a
semiconfined karst aquifer using time-lapse electrical resistivity tomography.
Water Resour. Res. 50, 2566–2585. https://doi.org/10.1002/2013WR013991
Mitchell, M.A., Oldenburg, D.W., 2016. Data quality control methodology for large ,
non-conventional DC resistivity datasets. J. Appl. Geophys. 3, 948–953.
https://doi.org/10.1016/j.jappgeo.2016.09.018
Morelli, G., LaBrecque, D.J., 1996. Advances in ERT inverse modeling. Eur. J. Environ.
Eng. Geophys. 1, 171–186.
References
157
Musgrave, H., Binley, A., 2011. Revealing the temporal dynamics of subsurface
temperature in a wetland using time-lapse geophysics. J. Hydrol. 396, 258–266.
https://doi.org/10.1016/j.jhydrol.2010.11.008
Nimmer, R.E., Osiensky, J.L., Binley, A.M., Sprenke, K.F., Williams, B.C., 2007.
Electrical resistivity imaging of conductive plume dilution in fractured rock.
Hydrogeol. J. 15, 877–890. https://doi.org/10.1007/s10040-007-0159-z
Parasnis, D.S., 1988. Reciprocity theorems in geoelectric and geoelectromagnetic work.
Geoexploration 25, 177–198. https://doi.org/10.1016/0016-7142(88)90014-2
Pearson, E.S., Hartley, H.O. (Eds.), 1970. Biometrika tables for statisticians. Cambridge
University Press.
Perri, M.T., Cassiani, G., Gervasio, I., Deiana, R., Binley, A., 2012. A saline tracer test
monitored via both surface and cross-borehole electrical resistivity tomography:
Comparison of time-lapse results. J. Appl. Geophys. 79, 6–16.
https://doi.org/10.1016/j.jappgeo.2011.12.011
Pinheiro, J., Bates, D., 1988. Unconstrained parameterizations for variance-covariance
matrices. Stat. Comput. 6, 289–296.
Ramirez, A., Daily, W., Binley, A., Labrecque, D., Roelant, D., 1996. Detection of leaks
in underground storage tanks using electrical resistance methods. J. Environ.
Eng. Geophys. 1, 189–203. https://doi.org/10.4133/JEEG1.3.189
Ramirez, A.L., Nitao, J.J., Hanley, W.G., Aines, R., Glaser, R.E., Sengupta, S.K., Dyer,
K.M., Hickling, T.L., Daily, W.D., 2005. Stochastic inversion of electrical
resistivity changes using a Markov Chain Monte Carlo approach. J. Geophys. Res.
110, B02101. https://doi.org/10.1029/2004JB003449
Revil, A., Johnson, T.C., Finizola, A., 2010. Three-dimensional resistivity tomography
of Vulcan’s forge, Vulcano Island, southern Italy. Geophys. Res. Lett. 37, 1–5.
https://doi.org/10.1029/2010GL043983
Robert, T., Dassargues, A., Brouyère, S., Kaufmann, O., Hallet, V., Nguyen, F., 2011.
Assessing the contribution of electrical resistivity tomography (ERT) and self-
potential (SP) methods for a water well drilling program in fractured/karstified
limestones. J. Appl. Geophys. 75, 42–53.
https://doi.org/10.1016/j.jappgeo.2011.06.008
Robinson, J., Johnson, T., Slater, L., 2015. Challenges and opportunities for fractured
rock imaging using 3D cross borehole electrical resistivity. Geophysics 80, E49–
E61. https://doi.org/10.1190/GEO2014-0138.1
References
158
Rubin, Y., Hubbard, S.S. (Eds.), 2005. Hydrogeophysics. Springer.
Schmidt-Hattenberger, C., Bergmann, P., Labitzke, T., Wagner, F., Rippe, D., 2016.
Permanent crosshole electrical resistivity tomography (ERT) as an established
method for the long-term CO2 monitoring at the Ketzin pilot site. Int. J. Greenh.
Gas Control 52, 432–448. https://doi.org/10.1016/j.ijggc.2016.07.024
Singha, K., Gorelick, S.M., 2006. Effects of spatially variable resolution on field-scale
estimates of tracer concentration from electrical inversions using Archie’s law.
Geophysics 71, G83. https://doi.org/10.1190/1.2194900
Slater, L., Binley, A., 2006. Synthetic and field-based electrical imaging of a zerovalent
iron barrier: Implications for monitoring long-term barrier performance.
Geophysics 71, B129–B137. https://doi.org/10.1190/1.2235931
Slater, L., Binley, A., 2003. Evaluation of permeable reactive barrier (PRB) integrity
using electrical imaging methods. Geophysics 68, 911–921.
https://doi.org/10.1190/1.1581043
Slater, L., Binley, A.M., Daily, W., Johnson, R., 2000. Cross-hole electrical imaging of a
controlled saline tracer injection. J. Appl. Geophys. 44, 85–102.
https://doi.org/10.1016/S0926-9851(00)00002-1
Slater, L.D., Ntarlagiannis, D., Day-Lewis, F.D., Mwakanyamale, K., Versteeg, R.J.,
Ward, A., Strickland, C., Johnson, C.D., Lane, J.W., 2010. Use of electrical imaging
and distributed temperature sensing methods to characterize surface water-
groundwater exchange regulating uranium transport at the Hanford 300 Area,
Washington. Water Resour. Res. 46, 1–13. https://doi.org/10.1029/2010WR009110
Uhlemann, S., Sorensen, J.P.R., House, A.R., Wilkinson, P.B., Roberts, C., Gooddy, D.C.,
2016. Integrated time-lapse geoelectrical imaging of wetland hydrological
processes. Water Resour. Res. 52, 1607–1625.
https://doi.org/10.1002/2015WR017932
Uhlemann, S., Wilkinson, P., Chambers, J., Maurer, H., Merritt, A., Gunn, D., Meldrum,
P., 2015. Interpolation of landslide movements to improve the accuracy of 4D
geoelectrical monitoring. J. Appl. Geophys. 121, 93–105.
https://doi.org/10.1016/j.jappgeo.2015.07.003
Vereecken, H., Binley, A., Cassiani, G., Revil, A., Titov, K. (Eds.), 2006. Applied
Hydrogeophysics, NATO Science Series. Springer Netherlands, Dordrecht.
https://doi.org/10.1007/978-1-4020-4912-5
References
159
Wagner, F.M., Möller, M., Schmidt-Hattenberger, C., Kempka, T., Maurer, H., 2013.
Monitoring freshwater salinization in analog transport models by time-lapse
electrical resistivity tomography. J. Appl. Geophys. 89, 84–95.
https://doi.org/10.1016/j.jappgeo.2012.11.013
Wallin, E.L., Johnson, T.C., Greenwood, W.J., Zachara, J.M., 2013. Imaging high stage
river-water intrusion into a contaminated aquifer along a major river corridor
using 2-D time-lapse surface electrical resistivity tomography. Water Resour. Res.
49, 1693–1708. https://doi.org/10.1002/wrcr.20119
Wehrer, M., Slater, L.D., 2015. Characterization of water content dynamics and tracer
breakthrough by 3-D electrical resistivity tomography (ERT) under transient
unsaturated conditions. Water Resour. Res. 51, 97–124.
https://doi.org/10.1002/2014WR016131
West, B., Welch, K.E., Galecki, A.T., 2007. Linear Mixed Models: a practical guide using
statistical software, 2nd ed. Chapman and Hall/CRC.
Wilkinson, P., Chambers, J., Uhlemann, S., Meldrum, P., Smith, A., Dixon, N., Loke,
M.H., 2016. Reconstruction of landslide movements by inversion of 4-D electrical
resistivity tomography monitoring data. Geophys. Res. Lett. 43, 1166–1174.
https://doi.org/10.1002/2015GL067494
Wilkinson, P.B., Chambers, J.E., Lelliott, M., Wealthall, G.P., Ogilvy, R.D., 2008.
Extreme sensitivity of crosshole electrical resistivity tomography measurements
to geometric errors. Geophys. J. Int. 173, 49–62. https://doi.org/10.1111/j.1365-
246X.2008.03725.x
Wilkinson, P.B., Chambers, J.E., Meldrum, P.I., Gunn, D.A., Ogilvy, R.D., Kuras, O.,
2010a. Predicting the movements of permanently installed electrodes on an
active landslide using time-lapse geoelectrical resistivity data only. Geophys. J.
Int. 183, 543–556. https://doi.org/10.1111/j.1365-246X.2010.04760.x
Wilkinson, P.B., Loke, M.H., Meldrum, P.I., Chambers, J.E., Kuras, O., Gunn, D.A.,
Ogilvy, R.D., 2012. Practical aspects of applied optimized survey design for
electrical resistivity tomography. Geophys. J. Int. 189, 428–440.
https://doi.org/10.1111/j.1365-246X.2012.05372.x
Wilkinson, P.B., Meldrum, P.I., Kuras, O., Chambers, J.E., Holyoake, S.J., Ogilvy, R.D.,
2010b. High-resolution Electrical Resistivity Tomography monitoring of a tracer
test in a confined aquifer. J. Appl. Geophys. 70, 268–276.
https://doi.org/10.1016/j.jappgeo.2009.08.001
Supplementary information
160
Williams, M.R., Buda, A.R., Singha, K., Folmar, G.J., Elliott, H.A., Schmidt, J.P., 2016.
Imaging Hydrological Processes in Headwater Riparian Seeps with Time-Lapse
Electrical Resistivity. Groundwater 1–13. https://doi.org/10.1111/gwat.12461
Yang, X., Chen, X., Carrigan, C.R., Ramirez, A.L., 2014. Uncertainty quantification of
CO2 saturation estimated from electrical resistance tomography data at the
Cranfield site. Int. J. Greenh. Gas Control 27, 59–68.
https://doi.org/10.1016/j.ijggc.2014.05.006
Yeh, T.-C.J., Zhu, J., Englert, A., Guzman, A., Flaherty, S., 2006. A successive linear
estimator for electrical resistivity tomography, in: Vereecken, H., Binley, A.M.,
Cassini, G., Revil, A., Titov, K. (Eds.), Applied Hydrogeophysics. Springer
Netherlands, pp. 45–74. https://doi.org/10.1007/978-1-4020-4912-5_3
Zhou, B., Dahlin, T., 2003. Properties and effects of measurement errors on 2D
resistivity imaging surveying. Near Surf. Geophys. 1, 105–117.
https://doi.org/10.3997/1873-0604.2003001
10) Supplementary information
LME electrode grouping formulation
To demonstrate the formulation of the new error model, we use a small subset
of the Boxford data to create different formulations of the linear mixed effects (LME)
error model. The first 25 measurements from the first set of Boxford are selected to fit
different models. We show that in Figure SI 1 that the linear or the binned linear
models perform poorly when fitting the reciprocal errors. By using LME models that
use average transfer resistance as the fixed effect, we see that much better fit of errors
is achieved by selecting any one of the four electrodes as a grouping variable. The LME
model essentially matches all the observed errors if all four electrodes are used as
grouping variables. It can be further improved by using dipole separation a and dipole-
dipole separation multiplier n as additional grouping variables.
We recommend fitting the LME error model in log-log scale to avoid negative
error estimates. Also, one should recognize if reciprocal errors are used, either the
configuration for “direct” and “reciprocal” set can be used to fit the model. Finally,
when there are incomplete reciprocal measurements, the LME model can be used in
“prediction mode” to estimate their errors.
Supplementary information
161
Fitting LME models to field data
We fit each of the 96 sets of Boxford reciprocal errors data individually with a
LME model, where average transfer resistance is the fixed effect variable and the 4
dipole-dipole electrodes (i.e. c1, c2, p1, p2) are the grouping variables. Figure SI 2
shows the evolution of the fixed-effect coefficients with time. The resultant coefficients
vary within a range but we can see they fluctuate significantly over time. The ±95%
confidence interval is also plotted as shaded regions. Figure SI 3 shows the random
effects coefficients of the resultant LME model for the first set of Boxford data as a
function of electrode number. We can see that the random effects coefficients span a
smaller range than the fixed effect ones. We can also see that the coefficients for the
four grouping variables (i.e. c1, c2, p1, p2) are somewhat correlated. It is important,
however, to note that random effect coefficients change discernibly with time.
We find that the binned linear model is quite stable at Sellafield regardless of
whether tracer was injected or not (Figure SI 4). It is therefore important to study
whether an LME model is stable too. Again, we fit each of the 246 sets of Sellafield data
individually with an LME model. Figure SI 5 shows the evolution of the fixed-effect
coefficients with time. Figure SI 6 shows the random effects coefficients of the resultant
LME model for the 5th February 2013 Sellafield data as a function of electrode number.
We can see that the coefficients are an order-of-magnitude higher than in the Boxford
data. We can also see that for certain electrode numbers the random effect coefficients
have much higher magnitudes, which indicates that some electrodes have a greater
impact on the observed errors from the fixed effect (i.e. transferred resistance). Finally,
the coefficients estimates do not show significant trends during the 2-year monitoring
periods, both during and outside the injection periods.
A guide to implement the LME error model in MATLAB®
To fit a LME model in MATLAB®, the statistics and machine learning toolbox
is required. Assume you have vectors of electrode configurations c1,c2,p1,p2, average
transferred resistances R, and observed measurement errors err, they first need to be
put into a MATLAB® table by using the command:
Supplementary information
162
tbl=table(log10(R),log10(err),c1,c2
p1,p2,'VariableNames','R','err','c1','c2','p1','p2');
Then fit the LME model with the following command:
lme = fitlme(tbl,'err ~ R+(R|c1)+(R|c2)+(R|p1)+(R|p2)');
This formulation assumes potential correlated random effects among grouping
variables c1, c2, p1, p2. To get the LME model-predicted measurement errors, use the
command:
fitted(lme)
If prediction of measurement errors is needed (i.e. the dataset contains measurements
with no error observations avaialable), use the command:
predict(lme,tblnew)
where tblnew is a table of vectors of electrode configurations where predictions are
made.
A guide to implement the LME error model in R
To fit a LME model in R, the lme4 package is required. Fit the LME error model
using the command:
lme <- lmer(Res_Err ~
Res_Ave+(Res_Err|C1)+(Res_Err|C2)+(Res_Err|P1)+(Res_Err|P2))
If you would like to run the above R script in Python, the rpy2
(https://rpy2.bitbucket.io/) interface may be used.
Supplementary information
163
Figure SI 1 Illustration plot to compare fitting linear and LME model for ERT measurement errors. The
LME error model provides a much better fit to measurement errors, especially the high measurement
errors.
Figure SI 2 Fixed effects coefficients for the LME error models for each of the Boxford datasets. The x-
axis is the dataset number in ascending order of collection (the time required to collect the 96 datasets
is around 24 hours). The left panel is when ‘intercept’ is used as the random-effect coefficient and the
right panel is for transfer resistance. The solid line is the maximum-likelihood estimate, while the
shaded area is enclosed by +/-95% confidence intervals.
Supplementary information
164
Figure SI 3 Random effects coefficients for the LME error model for a frame of the Boxford dataset.
The x-axis is the electrode number, where c1 and c2 are when the measurements are used as current
electrodes and p1 and p2 are when they are used as potential electrodes. The left panel is when
‘intercept’ is used as the random-effect coefficient and the right panel is for transfer resistance. The
solid line is the maximum-likelihood estimate, while the shaded area is enclosed by +/-95% confidence
intervals. Note the fixed effects for this linear mixed effect error model is 4.3708e-05 +/- 1.5628e-05 for
‘intercept’ and 1.5742e-04 +/- 7.527e-05 for transfer resistance.
Figure SI 4 Time series of linear error model for reciprocal errors at Sellafield. Errors higher than 20%
are not fitted. The blue line is the intercept of the linear model estimate, while the green and red lines
are the estimates for the slope. The three shaded regions are stimulant injection periods. It is shown
that the slope of the linear error models (i.e. percentage errors) did not vary significantly during the
monitoring period and the stimulant injection has little impact on error levels.
Supplementary information
165
Figure SI 5 Time series of the fixed effect coefficients estimates for the LME error model for Sellafield.
The three shaded regions are stimulant injection periods.
Figure SI 6 Random effect coefficients for LME error model at Sellafield on 5th February, 2014. The x-
axis is the electrode number, where c1 and c2 are when the measurements are used as current electrodes
and p1 and p2 are when they are used as potential electrodes. The left panel is when ‘intercept’ is used
as the random-effect coefficient and the right panel is for transfer resistance. The solid line is the
maximum-likelihood estimate, while the shaded area is enclosed by +/-95% confidence intervals. Note
the fixed effects for this linear mixed effect error model is 0.47852 +/- 0.4865 for ‘intercept’ and 0.01428
+/- 0.027414 for transfer resistance.
Paper 2: Integrated hydrogeophysical modelling and data assimilation for
geoelectrical leak detection
166
5. Paper 2: Integrated hydrogeophysical modelling and data
assimilation for geoelectrical leak detection
Submitted for peer review to Journal of Contaminant Hydrology as Tso, C.-H.M.,
Johnson, T., Song, X., Chen, X., Kuras, O., Wilkinson, P., Uhlemann, S., Chambers, J.,
Binley, A. (201x) Integrated hydrogeophysical modelling and data assimilation for
geoelectrical leak detection
Abstract
167
Abstract
Time-lapse electrical resistivity tomography (ERT) measurements provide indirect
observations of hydrological processes in the Earth's shallow subsurface at high spatial
and temporal resolution. ERT has been used in the past decades to detect leaks and
monitor the evolution of associated contaminant plumes. Specifically, inverted
resistivity images allow visualization of the dynamic changes in the structure of the
plume. However, existing methods do not allow the direct estimation of leak
parameters (e.g. leak rate, location, etc.) and their uncertainties. We propose an
ensemble-based data assimilation framework that evaluates proposed hydrological
models against observed time-lapse ERT measurements without directly inverting for
the resistivities. Each proposed hydrological model is run through the parallel coupled
hydro-geophysical simulation code PFLOTRAN-E4D to obtain simulated ERT
measurements. The ensemble of model proposals is then updated using an iterative
ensemble smoother. We demonstrate the proposed framework on synthetic and field
ERT data from controlled tracer injection experiments. Our results show that the
approach allows joint identification of contaminant source location, initial release time,
and solute loading from the cross-borehole time-lapse ERT data, alongside with an
assessment of uncertainties in these estimates. We demonstrate a reduction in site-
wide uncertainty by comparing the prior and posterior plume mass discharges at a
selected image plane. This framework is particularly attractive to sites that have
previously undergone extensive geological investigation (e.g., nuclear sites). It is well
suited to complement ERT imaging and we discuss practical issues in its application
to field problems.
Graphical abstract
168
Graphical abstract
Estimation of leak parameters and their uncertainties using raw geophysical data and data
assimilation.
1) Introduction
Identification of solute loadings from an unknown source is a complex yet
critical problem. For example, understanding the whereabouts of the source(s) of
contamination is often the first question that needs to be addressed in a remediation
project. This identification, however, is not straightforward and it is often complicated
by factors such as unknown forcing (e.g., boundary and flow conditions), aquifer and
vadose zone heterogeneity, and limited data (in terms of number, types, temporal and
spatial coverage). Because of these complications, attempts to assess source
identification should also address the uncertainties in the estimates, and provide
realistic and actionable uncertainty bounds.
Traditional point-based sampling methods suffer from limited coverage and
resolution. As prompted, in part, by the wealth of studies in stochastic subsurface
hydrology that argued for better field techniques, geophysical methods have emerged
as valuable tools for investigating shallow subsurface processes over the past two
decades (Binley et al., 2015). Geophysical methods can provide much larger spatial and
temporal resolution. Once installed, autonomous long-term monitoring systems, such
Introduction
169
as ALERT (Kuras et al., 2009), can repeatedly collect geophysical data and transmit it
back to the office using telemetry. Among them, electrical resistivity tomography (ERT)
is particularly suitable for leak detection due to its sensitivity to fluid conductivity.
Note that leak detection is not limited to the detection of the breakthrough of saline
fluids (as proxies of contaminants), but it also includes monitoring the integrity of
water-retaining structures (e.g. embankments or levees) (Abdulsamad et al., 2019) and
landfills (Audebert et al., 2014; Chambers et al., 2006; Maurya et al., 2017).
Many previous ERT studies have focused on inferring plume characteristics by
delineating the plume geometry (Aghasi et al., 2013), obtaining summary statistics of
the plume structure (e.g. spatial and temporal movements) (Crestani et al., 2015;
Pidlisecky et al., 2011; Singha and Gorelick, 2006), or developing methods for
automatic tracking of plumes (Ward et al., 2016). There is also a substantial amount of
work dedicated to delineating local hydraulic properties using ERT (e.g. Camporese et
al., 2011). As an effort to better use geophysical data for hydrogeological studies,
comparisons between coupled and uncoupled hydrogeophysical inversions of ERT-
monitored tracer tests have been made (Camporese et al., 2015; Hinnell et al., 2010).
Others have tried to address the uncertain link between hydrological systems and
geophysical data using data-driven or machine learning approaches (Hermans et al.,
2016b, 2015; Oware et al., 2013). There is also increased use of geophysics to estimate
remediation efficiency (LaBrecque et al., 1996b). For example, Power et al. (2014)
applied 4D active time-constrained inversion to time-lapse ERT data to estimate the
volume of solute plume remediated in a laboratory experiment.
The various electrical methods applied to the mapping and monitoring at the
U.S. Department of Energy Hanford nuclear site has greatly improved the readiness of
these methods (Johnson et al., 2015a). For example, the work on the monitoring of the
groundwater/river water interaction beneath the Hanford 300 Area infiltration bonds
(Johnson et al., 2012; Johnson et al., 2015b; Slater et al., 2010; Wallin et al., 2013) shows
ERT is well suited for monitoring such complex and dynamic processes, while the
successful monitoring of vadose zone desiccation (Truex et al., 2013b, 2012) at the BC
Cribs Area demonstrates its capability to monitor 3-D changes in moisture content
Introduction
170
caused by gas injection. The leak tank experiments in the 1990s and 2000s have
contributed some important work in geoelectrical leak detection. The first two mock
tank experiments set up a 15 m diameter steel tank at the Hanford site and ERT
tomograms clearly shows area of resistivity decrease of the leak plume (Ramirez et al.,
1996). A subsequent series of mock tank experiments evaluated a number of electrical
methods for leak detection (Barnett et al., 2003). Among them, a “blind test” was
carried out for 110 days where the release episodes were not known to the modeller
(Daily et al., 2004). The modeller achieved a 57% success rate in defining a leak or no
leak declaration during the test, although further analysis have greatly improved the
success rate. A follow-up study on the dataset used Markov chain Monte Carlo
inversion to estimate the probability distribution of the plume of being in different
sizes and shapes (Ramirez et al., 2005).
In groundwater hydrology or hydrogeophysical problems, models are often
too complex (in terms of parameterisation) such that fully Bayesian methods such as
Markov chain Monte Carlo (McMC) methods are rarely applied (Irving and Singha,
2010). Data assimilation has played an increasingly important role in subsurface
characterization (Zhou et al., 2014). For example, Chen et al. (2013) used p-space
ensemble Kalman filter (EnKF) (Nowak, 2009; Schöniger et al., 2012) and ensemble
smoother (ES) to assimilate head, flowmeter, and conservative tracer test data to
characterize the permeability field of the Hanford 300 area. Zovi et al. (2017) used
surface ERT results to generate facies model that honour the geophysical data, then
used restart normal-score EnKF to estimate the hydraulic conductivity (K) field. In a
recent review, it was concluded that the iterative ES (IES) could achieve results
comparable with those of the EnKF, at a fraction of EnKF’s computational cost (Li et
al., 2018). This computational saving stems from the difference in their formulation—
in the EnKF, the data are sequentially integrated into the model at simulation time
steps while in ES all the data are combined together and assimilated only once (note in
IES the amount of data between updating steps are the same). Since EnKF assimilates
data in a sequential fashion (i.e. one time step after another), the number of
Introduction
171
assimilation steps equals the number of time steps present in the data. Therefore, EnKF
is more computationally expensive than IES when data from many time steps are used.
The Hanford leak tank studies and other earlier work on geoelectrical leak
monitoring have focused on obtaining time-lapse ERT images during the suspected
leak, and making “leak/ no leak” decisions based on the images. It is difficult, however,
to use geophysical images to infer leak parameters such as leak location, solute loading,
and onset time. Recent hydrogeophysical studies have attempted to estimate
parameters of interest from geophysical data without inverting for geophysical images.
Different hydrological model proposals are evaluated and compared to observed
geophysical data. For example, Manoli et al. (2015) used an iterative particle filter
approach and a coupled hydrogeophysical forward model to estimate hydraulic
conductivity, K, of up to four zones from ERT data obtained during a controlled
infiltration experiment. This approach is then extended to a field study which
considers both ERT and ground penetrating radar (GPR) data in K estimation (Rossi et
al., 2015). Scholer et al. (2012) used time-lapse crosshole ground GPR data collected
under different infiltration conditions to estimate unsaturated soil hydraulic properties
using a McMC inversion. Kowalsky et al. (2005) jointly estimated the dielectric and
unsaturated zone parameters using both GPR and hydrological data. Johnson et al.
(2009) developed a data-domain correlation approach for joint hydrogeological
inversion of time-lapse hydrogeological and ERT data to jointly estimate fluid solute
concentration and resistivity without explicitly specifying a petrophysical transform.
Though contaminant source identification has been a persistent problem in
hydrogeology (Michalak and Shlomi, 2007; Shlomi and Michalak, 2007; Sun, 2007; Sun
et al., 2006; Sun and Sun, 2015), advances in data assimilation methods have opened a
new avenue in addressing this problem. Only a few studies have jointly estimated leak
parameters and hydraulic parameters (Datta et al., 2009; Koch and Nowak, 2016;
Wagner, 1992). Zeng et al. (2012) developed a sparse grid Bayesian method for
contaminant source identification, which greatly reduced the computational burden in
McMC sampling and accurately identifies both leak parameters and time-varying
source strengths in case studies. Xu et al. (2016) simultaneously identified the above
Introduction
172
contaminant source parameters using the restart normal-score ensemble Kalman filter,
while subsequently Xu et al. (2018) extended the method to also identify the
heterogeneous hydraulic conductivity field. The method has recently been applied to
a sandbox study (Chen et al., 2018), where six leak parameters and 2 parameters for
the location of an impermeable plate are estimated. Assuming known source location,
Kang et al. (2018) estimated K and Dense Non-Aqueous Phase Liquid (DNAPL)
saturation (and thus total DNAPL volume) from ERT data using restart EnKF.
In contaminated land studies, there has been a paradigm shift to focus more on
site-wide metrics. Instead of focusing on thresholds from point-based measurements,
mass discharge and mass flux has been used increasingly (Brusseau and Guo, 2014;
Christ et al., 2010, 2006; Hadley and Newell, 2012). Several studies are dedicated to
studying their estimation and uncertainty bounds from point measurements (Cai et al.,
2011; Troldborg et al., 2012, 2010), while Balbarini et al. (2018) used regression kriging
of collocated concentration and geoelectrical data to improve mass discharge estimates.
In this paper, we introduce an ensemble-based data assimilation framework to
jointly identify various leak parameters with their associated uncertainty bounds from
ERT data. The method evaluates proposed hydrological models (i.e. different
hydrogeological units, different leak locations and loads) against observed time-lapse
ERT measurements. To the best of our knowledge, this work is the first attempt to
estimate solute source parameters using raw ERT data, as most previous work focuses
on estimating hydraulic parameters or reconstructing solute distribution. A key
feature of our method is that it allows visualization of uncertainty reduction by
comparing the envelopes of prior and posterior mass discharge curves. The methods
and data used in this work are detailed in section 2. Results of the various synthetic
and field test cases are reported in section 3 and 4 respectively. Finally, we discuss and
summarize our findings in section 5 and 6 respectively.
Methodology
173
2) Methodology
We begin by outlining the different steps in the framework, followed by details
of the different framework components. Finally, we introduce the datasets used in test
cases.
2.1. Overview of framework
The data assimilation framework (summarized by Figure (a)) begins by
proposing a range of hydrological models (i.e. model parameters such as leak
locations). All parameters for variably saturated flow and transport simulation need to
be prescribed, either as a fixed constant or a distribution (which will be updated by the
DA framework). Also, the setup for the ERT experiment (e.g. mesh, electrode locations,
measurement protocols, petrophysical transforms) need to be included. Once we have
an ensemble of model proposals, they are fed to simulate the ERT response using
PFLOTRAN-E4D (Johnson et al., 2017). The misfits between observed and simulated
ERT responses are used to form data error covariance matrices, which in turn are used
to update the model proposals. The entire process repeats until the misfit criterion is
met or the algorithm reaches the user-specified maximum number of iterations.
Figure 1 (a) Flowchart of the overall data assimilation framework used in this work. More details are
found in the subsections. (b) The goal of this framework is that upon conditioning of geophysical data,
the envelope of possible mass discharge time series will become less uncertain.
Methodology
174
2.2. Coupled hydrogeophysical forward modelling
We use the massively parallel code PFLOTRAN-E4D (Johnson et al., 2017) for
coupled hydrogeophysical forward modelling. E4D (Johnson et al., 2010) is an ERT
code which has state-of-the-art capability for parallelization and for accurate
modelling of metallic infrastructure (e.g. tanks and pipes that are common at
contaminated sites) (Johnson and Wellman, 2015) and near-real-time inversion to
monitoring bioremediation (Johnson et al., 2015c). E4D has been used for ERT
modelling on a number of complex problems such as those at the Hanford Site.
PFLOTRAN (Hammond and Lichtner, 2010 , also see pflotran.org) is a state-of-the-art
massively parallel subsurface flow and reactive transport code. PFLTORAN-E4D
(implemented as “hydrogeophysics” mode in the 2018 PFLOTRAN distributions used
in this work) translates states of the PFLTORAN model to bulk electrical conductivity
𝜎𝑏 distribution using an interpolation matrix that maps between the meshes of the two
codes given a petrophysical transform. To do so, users need to provide elementwise
petrophysical parameters (e.g. Archie parameters), times when the simulated ERT
measurements are needed, and the fluid conductivities of the groundwater and the
injected tracer. In this work, we assume surface electrical conductivity is negligible and
use Archie’s law as the petrophysical relationship:
𝜎𝑏 = 𝜎𝑤𝛷𝑚𝑆𝑤
𝑛 (1)
where 𝑚, is the cementation exponent, and 𝑛 is the saturation exponent. Specifically,
fluid conductivity 𝜎𝑤 , porosity 𝛷 , and fluid saturation 𝑆𝑤 are passed from the
PFLOTRAN output to E4D through the mapping routine. After the petrophysical
mapping, E4D will run a forward simulation with the given ERT survey configuration
and 𝜎𝑏 distribution to produce the simulated ERT data. Note that PFLOTRAN-E4D is
no longer supported in newer PFLOTRAN releases. The mapping routine is available
through the corresponding author.
2.3. Prior parameter generation: Latin hypercube sampling
For multi-parameter data assimilation problems, we need to use an efficient scheme to
generate 𝑛𝑟𝑒𝑎𝑧 model proposals. We use Latin hypercube sampling (LHS) to obtain
multi-parameter model proposals that efficiently span the parameter space. The LHS
Methodology
175
approach is implemented using the R package Envstats (Millard, 2013). For the
synthetic and field examples, we assume multivariate Gaussian distribution (𝑛𝑟𝑒𝑎𝑧 =
32) and multivariate uniform distribution (𝑛𝑟𝑒𝑎𝑧 = 64) for the prior distribution of
parameter values respectively. The use of more realizations and a non-informative
prior in the field example is due to greater parameter uncertainty.
2.4. Data assimilation: ensemble smoother with multiple data assimilation (ES-
MDA)
In this work, we use the ensemble smoother with multiple data assimilation
(ES-MDA) (Emerick and Reynolds, 2013) to update hydrological models. ES-MDA is
also known as an iterative variant of ensemble smoother (ES). The ES-MDA has been
used heavily in hydrocarbon reservoir history matching of production and seismic
data, but there are growing applications in hydrology. For example, Ju et al. (2018)
combined ES-MDA with Gaussian process surrogate modelling and tested the new
method on synthetic 2-D transient groundwater flow problems. Lan et al. (2018)
combined sequential ensemble-based optimal design and ES-MDA to accurately and
efficiently estimate the heterogeneous distribution of physical and geochemical
parameters in groundwater models. Aalstad et al. (2018) used ES-MDA and fractional
snow-covered area retrieved from satellites to estimate the snow distribution at Arctic
sites. Song et al. (2019) used ES-MDA with level set parameterization to estimate the
three-facies heterogeneous permeability field at the Hanford IFRC site, while Kang et
al. (2019) jointly assimilated ERT and concentration data using ES-MDA alongside
with direct sampling (Mariethoz et al., 2010) to estimate the non-Gaussian hydraulic
conductivity field from a synthetic salt injection experiment. More recently, a modified
version of ES-MDA has been used for crosshole GPR travel-time tomography in
conjunction with approximate forward solvers and model error correction (Köpke et
al., 2019).
An ensemble smoother (ES) considers all available time-lapse data
simultaneously for updating the model parameters. The ES-MDA method essentially
allows iterative updating of the nonlinear ES problem by inflating the observational
errors by a factor α and solve the updating equation α times iteratively. It has been
Methodology
176
shown that iterative updating better handles nonlinearity in the data assimilation
problem than the classic ES formulation. Our implementation of the ES-MDA
procedure is summarized below:
1. Prepare observational data (and their error levels) to be used for data assimilation
(DA)
2. Set up a base PFLOTRAN-E4D model
3. Decide on which parameter(s) to update, either based on expert judgement or some
preliminary global sensitivity analysis. The parameter estimation may be affected
if important parameters are neither assumed correctly nor updated. Sample 𝑁𝑒
realizations from the prior distribution of parameter(s) values (e.g. assume normal
or uniform distribution) to obtain parameter array m at 𝑙 = 0 (𝒎0). Parameters
that are not being updated are assumed known and base model values are used
throughout the DA process for all realizations.
4. Run PFLOTRAN-E4D using 𝒎0 to obtain an ensemble of simulated ERT data
5. Updating. For 𝑙 = 1 to 𝑁𝑎 (where 𝑁𝑎 is the number of data assimilation steps),
(i.) The data misfit from the (𝑙 − 1)-th iteration is given by
𝑚𝑖𝑠𝑓𝑖𝑡 =∑ ∑ (𝑑𝑖,𝑜𝑏𝑠−𝑑𝑖,𝑗)
𝑁𝑒𝑗=1
𝑁𝑑𝑖=1
𝑁𝑒× 𝑁𝑑 (2)
where 𝑁𝑑 is the number of measurements and 𝑑𝑖,𝑗 is the 𝑖-th data of the
𝑗-th realization.
(ii.) Obtain the auto covariance matrix of model predictions 𝑪𝐷𝐷 and the
cross-covariance matrix between the parameter vector and model
predictions 𝑪𝑀𝐷 by
𝑪𝐷𝐷 = cov(𝒅𝑗, 𝒅𝑗) ≈
1
𝑁𝑒−1∑ (𝒅𝑗 − )(𝒅𝑗 − )
𝑇𝑁𝑒𝑗=1 (3)
𝑪𝑀𝐷 = cov(𝒎𝑗, 𝒅𝑗) ≈
1
𝑁𝑒−1∑ (𝒎𝑗 − )(𝒅𝑗 − )
𝑇𝑁𝑒𝑗=1 (4)
where 𝒅𝑗 and 𝒎𝑗 are vectors of simulated data and model parameter
estimates of the 𝑗-th realization, respectively. The overbar denotes the
mean across realizations of a matrix.
Methodology
177
(iii.) For each ensemble member, perturb the observation vector using
𝒅𝑢𝑐 = 𝒅𝑜𝑏𝑠 +√𝛼𝑙𝑪𝐷1/2𝒛𝑑 (5)
where 𝛼𝑙 is an inflation coefficient, 𝒛𝑑~𝑁(0, 𝑰𝑁𝑑) , 𝑰𝑁𝑑 is an identity
matrix of size 𝑁𝑑, 𝑪𝐷 is the covariance matrix of the measurements error,
𝒅𝑜𝑏𝑠 is a vector of the observed field data. Because of the linear
approximation in the update step and the use of a limited number of
realizations in an ensemble, the ensemble Kalman filter-like methods
have a tendency to systematically underestimate the variance of the
model variables (Zhang et al., 2010). Resampling the vector of perturbed
observations at each iteration tends to reduce sampling problems
caused by matching outliers that may be generated when perturbing the
observations (Emerick and Reynolds, 2013).
(iv.) Update the parameter ensemble using:
𝒎𝑙 = 𝒎𝑙−1 + 𝑪𝑀𝐷(𝑪𝐷𝐷 + 𝛼𝑙𝑪𝐷)−1⏟
𝐾𝑎𝑙𝑚𝑎𝑛 𝑔𝑎𝑖𝑛
(𝒅𝑢𝑐 − 𝒅𝑙−1)𝑇⏟
𝑚𝑖𝑠𝑓𝑖𝑡
(6)
Note that in order to preserve the equivalence between single and
multiple data assimilation, it is necessary that ∑ 1/𝛼𝑙𝑁𝑎𝑙 = 1 (Emerick
and Reynolds, 2013). This effectively serves to update the average
sensitivity matrix.
(v.) Run PFLOTRAN-E4D 𝑁𝑒 times using 𝒎𝑙 to obtain the updated
simulated data ensemble
7. If solution does not converge, repeat steps 3-7 with a higher α and/or 𝑛𝑟𝑒𝑎𝑧.
Convergence is achieved when the ensemble root-mean-square-error of the
ERT data misfit equilibrates:
𝑅𝑀𝑆𝐸 = √1
𝑁𝑑
1
𝑁𝑒∑ ∑ (𝑑𝑖
𝑜𝑏𝑠 − 𝑑𝑖,𝑗𝑠𝑖𝑚)
𝑁𝑒𝑗=1
𝑁𝑑𝑖=1 (7)
ES-MDA outperforms ES in non-linear problems because the smoother
effectively represents a single Gauss–Newton iteration with a full step and an average
sensitivity matrix (Reynolds et al., 2006) that is approximated by the covariance
matrices of the prior ensemble. Instead of a single and potentially large Gauss-Newton
correction, ES-MDA allows multiple smaller corrections through the use of multiple
Methodology
178
iterations. To damp the parameter updating and to correctly sample from the posterior
distribution of model parameters, the covariance matrices must be inflated (Emerick
and Reynolds, 2013). The above issue is sometimes referred to as residual sampling
error in data assimilation literature, with inflation being a common technique to
account for it (Carrassi et al., 2018). The ES-MDA scheme is more flexible and easier to
implement than conventional Gauss-Newton methods because it does not require
derivation of sensitivity matrices. Previous work have shown that good results can be
obtained in a few iterations (e.g. 4-10), while using a decreasing order of 𝛼𝑙‘s only
resulted in small improvements compared to using constant 𝛼𝑙‘s. It can be shown that
ES-MDA has links to annealed importance sampling (Stordal and Elsheikh, 2015). A
comparison between ES-MDA and ES is provided by Evensen (2018).
In this work, the above steps (except forward modelling) were implemented in
R. For the synthetic studies presented, we set 𝑁𝑎 to 7 and use a constant 𝛼𝑙 of 7, which
appears to obtain convergence in all cases and also satisfies the criterion ∑ 1/𝛼𝑙𝑁𝑎𝑙 = 1.
Because the initial misfit for the field data is much larger than that for the synthetic
data, the algorithm was unstable and more difficult to converge. Thus, for our field
study we set a constant 𝛼𝑙 to 200 and iterate until the RMSE is stabilized, which is
achieved within ten iterations. Although this violates the ∑ 1/𝛼𝑙𝑁𝑎𝑙 = 1 criterion, we
remark that its choice is determined based on data noise levels and discrepancy
between observed and simulated data, which can be high in field data. A higher 𝛼𝑙 can
be seen as adding regularization to the ensemble Kalman scheme (Iglesias, 2016). An
alternative approach is to adaptively decide 𝛼𝑖 at each iteration automatically (e.g.
Iglesias and Yang, n.d.; Le et al., 2016) based on the mean of RMSE of data misfit across
all realizations.
In previous hydrogeology applications using ensemble Kalman methods, the
hydraulic heads or solute concentrations are often transformed using normal-score
transformation (e.g. Schöniger et al., 2012). We consider ERT data to be more Gaussian
than hydrogeological data so we use raw ERT data (transfer resistances) directly in this
study but such scaling may improve results.
Synthetic experiments based on the Sellafield ERT field trial
179
2.5. Plume mass discharge
Mass discharge is the integral of solute fluxes across a control plane (ITRC,
2010). The control plane can be a model or site boundary, the water table, or any
arbitrary planes. Mass flux is defined as 𝐽 = 𝑞0𝐶, where is 𝑞0 groundwater flux and 𝐶
is solute concentration. It follows that the solute mass discharge (or equivalently solute
integral flux) across a control plane is defined as 𝑀𝑑 = ∫ 𝐽𝑑𝐴𝐴
, where 𝐴 is area of the
control plane and 𝐽 is the spatially variable solute mass flux. Note that since the solute
fluxes are vectors, it is possible for solute mass discharge to be negative. As shown in
Figure 1(b), one way to visualize reduction in site-wide uncertainty is by observing a
reduction of spread of the mass discharge time series.
3) Synthetic experiments based on the Sellafield ERT field trial
Between 2013-2014, a field ERT trial was conducted at the Sellafield Nuclear
Site in Cumbria, U.K. (Kuras et al., 2016; Tso et al., 2017) by the British Geological
Survey to demonstrate the utility of a permanent ERT monitoring system to support
critical decommissioning activities at nuclear sites. Four vertical boreholes and two
inclined boreholes with forty electrodes each were installed in front of the Sellafield
MSSS building. The field trial included three controlled injections of an electrically
conductive tracer (as simulant of the silo liquor) into the vadose zone. Time-lapse ERT
data were collected during the experiment.
We built a PFLOTRAN model based on the hydrogeological model developed
for Sellafield (Kwong and Fowler, 2014) and an E4D model based on the electrode
locations and design of the field trial. Details of the PFLOTRAN and E4D models are
found in Table 2. Note that there are multiple units in the domain, but only the
hydraulic parameters in the main unit (i.e. sandy drift) is listed in Table 2. The
parameters not being estimated are kept constant during parameter estimation.
To test our method, we obtained synthetic ERT data based on the experimental
setup of the field trial and consider a series of parameter estimation cases. They are
summarized in
Synthetic experiments based on the Sellafield ERT field trial
180
Table 3. Unless otherwise stated, the parameters not being estimated are
assumed to be known exactly. We began by considering the estimation of leak location
(𝑥𝑙𝑜𝑐 ,𝑦𝑙𝑜𝑐 ), both for a leak inside and outside the ERT monitoring cell. Then we
proceed by also estimating the solute loading (𝑞), release onset time (𝑡0). Subsequently,
we estimate both leak parameters and uniform Archie parameters (𝑚, 𝑛) jointly, which
is important in field applications as fixing the parameters imposes too much
confidence on uncertain petrophysical relationships. Finally, we consider a few cases
with uncertainty and heterogeneity in hydraulic conductivity (𝐾). In the first case, the
𝐾 field has a log variance of 1.0 but its mean value is unknown; while in the second
case, the 𝐾 field is heterogeneous but its mean value is known. In the last case, the
mean 𝐾 value is being estimated for a heterogeneous field. Other potential parameters
to consider includes water table depths, permeability [log10 (m2)], porosity,
unsaturated zone van Genuchten parameters, recharge rates, depth of the leak (𝑧𝑙𝑜𝑐),
and duration of the leak (𝑑𝑡). Each iteration takes 40 minutes on average to run on 192
cores on PNNL’s institutional computing facility. Note that only the forward
modelling is parallelized, not the parameter updating.
Figure 2 . (a) PFLOTRAN model domain for the Sellafield MSSS. The grey area is the MSSS building,
which is modelled as impermeable. The hashed area is the ERT imaging cell consisting of four ERT
boreholes. (b) A snapshot of the simulated tracer concentration due to injection. (c) The corresponding
distribution of electrical conductivity within the ERT imaging cell obtained via petrophysical
transform.
Synthetic experiments based on the Sellafield ERT field trial
181
Table 2 “True” coupled hydrogeophysical model parameters used for synthetic experiments. It is
developed based on the Sellafield field trial. *Only parameters for the main zone are listed below.
#Leak location for some cases is (33.4534, -14.4303) instead. Note that for all cases the leak location is
at the water table.
PFLOTRAN simulation Value
Total simulation time (days) 30
Model dimensions (m) 40 x 40 x 20
Grid spacing (m) 1 x 1 x 1
Horizontal permeability (m2) * 8.8854 x 10-10
Vertical permeability (m2) * 4.4427 x 10-11
Porosity * 0.2
Water table depth (m) 6.0
van Genuchten 𝑚 0.5
van Genuchten 𝛼 1 x 10-4
Residual water saturation 0.1
Leak location (m) # (20,-10,18.1)
Leak period (day) 12-30
Leak rate (m3/d) 8.0
Background fluid conductivity (S/m) 1 x 10-4
Leak fluid conductivity (S/m) 0.1
Mass discharge plane Vertical plane at y=-25.03m
E4D simulation Value
Full Model dimensions (m) 100 x 100 x 100
Imaging cell dimensions (m) 9.5 x 22.8 x 41.5
Grid spacing Unstructured
Number of elements 380457
ERT imaging times (day) Every 5 days between day 5 to day 30
Archie’s cementation exponent 1.3
Archie’s saturation exponent 2.0
Synthetic experiments based on the Sellafield ERT field trial
182
Table 3 Summary of synthetic cases. All cases converge in seven iterations.
Figure
Size of
ensemble
(𝑁𝑒)
Parameter(s)
to estimate Prior distribution Comments
Initial and
final RMSE
Figure 3a 32 𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐 6x6 grid (exclude corners)
Uniform spacing
X range: -5 – 55m
Y range: -33 - -3m
Estimation of the
leak location on
the x,y plane ; leak
is located within
the ERT cell
3.63 1.01
Figure 3b 32 𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐
6x6 grid (exclude corners)
Uniform spacing
X range: -5 – 55m
Y range: -33 - -3m
Estimation of the
leak location on
the x,y plane; leak
location is outside
the ERT cell
7.66 1.01
Figure 4 32 𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐,𝑞, 𝑡0
Multivariate uncorrelated truncated Gaussian:
xloc = list(mean=25.0, sd=20.0, min=-
5.0,max=55.0),
yloc = list(mean=-18.0,sd=10.0,min=-33.0,
max=-3.0),
q = list(mean=15.0,sd=10.0,min=0.0, max=30.0),
t0 = list(mean=15.0,sd=10.0,min=0.0, max=30.0)
Estimation of the 4
leak parameters
7.01 1.01
Figure 5 32 𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐,𝑞,
𝑡0, 𝑚, 𝑛
Multivariate uncorrelated truncated Gaussian:
xloc = list(mean=25.0, sd=20.0, min=-
5.0,max=55.0),
yloc = list(mean=-18.0,sd=10.0,min=-33.0,
max=-3.0),
q = list(mean=15.0,sd=10.0,min=0.0, max=30.0),
t0 = list(mean=15.0,sd=10.0,min=0.0, max=30.0),
c = list(mean=1.6,sd=0.5,min=0.0, max=2.0),
m = list(mean=2.5,sd=0.8,min=0.0, max=3.0)
Joint estimation of
leak parameters
and uncertain
(homogeneous)
petrophysical
parameters
(Archies
cementation factor
and saturation
exponent)
22.65 1.65
Figure 6a 32 𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐,𝑞, 𝑡0
Multivariate uncorrelated truncated Gaussian:
xloc = list(mean=25.0, sd=20.0, min=-
5.0,max=55.0),
yloc = list(mean=-18.0,sd=10.0,min=-33.0,
max=-3.0),
q = list(mean=15.0,sd=10.0,min=0.0, max=30.0),
t0 = list(mean=15.0,sd=10.0,min=0.0, max=30.0)
Leak estimation
under the
influence of
permeability
heterogeneity
3.30 1.10
Figure 6b 32 𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐,𝑞, 𝑡0
Multivariate uncorrelated truncated Gaussian:
xloc = list(mean=25.0, sd=20.0, min=-
5.0,max=55.0),
yloc = list(mean=-18.0,sd=10.0,min=-33.0,
max=-3.0),
q = list(mean=15.0,sd=10.0,min=0.0, max=30.0),
t0 = list(mean=15.0,sd=10.0,min=0.0, max=30.0)
Leak estimation
under the
influence of and
uncertain
(homogeneous)
permeability
6.47 1.31
6.66 1.70
6.50 1.22
6.53 1.41
6.54 1.27
Figure 7 32 𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐,𝑞,
𝑡0, 𝐾
Multivariate uncorrelated truncated Gaussian:
xloc = list(mean=25.0, sd=20.0, min=-
5.0,max=55.0),
yloc = list(mean=-18.0,sd=10.0,min=-33.0,
max=-3.0),
q = list(mean=15.0,sd=10.0,min=0.0, max=30.0),
t0 = list(mean=15.0,sd=10.0,min=0.0, max=30.0),
K = list(mean=-9.0,sd=sqrt(1.0),min=-11.0,
max=-7.0))
Joint estimation of
leak parameters
and uncertain
(homogeneous)
permeability
values
3.30 1.03
Synthetic experiments based on the Sellafield ERT field trial
183
3.1. Base cases
Our initial example considers the estimation of the leak location (Figure 3). The
prior realizations are laid in a rectangular grid. We consider both the cases where the
leak is within and outside the ERT imaging cell. Although the estimate at the first
iteration is superior when the leak is within the imaging cell, the leak location is
accurately estimated after seven iterations in both cases. Figure 4 shows the results
from the joint estimation of four leak parameters: the (𝑥, 𝑦) coordinates of the leak
location, leak rate, and onset time, assuming a wide multivariate Gaussian prior
distribution. After conditioning the parameter values with ERT data, all four leak
parameters are accurately estimated. Figure 4b shows the mass discharge curves across
a pre-defined plane. The mass discharge curves for the prior distribution are highly
variable, while those for posterior distribution collapse to the true curve.
Figure 3 Estimation of leak location. (a) The true leak location is within the ERT array (33.4534, -
14.4303). (b) The true leak location is outside the ERT array (20, -10). In both cases, the data
assimilation framework successfully identified the true leak location within a few iterations.
Synthetic experiments based on the Sellafield ERT field trial
184
Figure 4 Joint estimation of leak parameters: (𝒙, 𝒚) location, leak rate, and onset time. (a) Violin plots
showing the prior and posterior parameter distributions. The true values are marked with an orange
lines. The posterior parameter values collapse around the true values (b) Prior and posterior tracer
mass flux across the pre-defined plane. All the posterior curves collapse to nearly the true curve (green).
Note that the sign of mass discharge denotes its direction across the plane.
3.2. Effects of petrophysical parameters
Figure 5 shows the joint estimation of leak parameters and Archie
petrophysical parameters. The prior estimates are generated as multivariate
Gaussian distributions using Latin hypercube sampling. The posterior
estimates are in very good agreement with the true values, with the exception
that the onset time is slightly underestimated. It is noteworthy that including
the Archie parameters as a covariate has caused the RMSE of the prior
ensemble to be much higher than those in other synthetic test cases (see
Synthetic experiments based on the Sellafield ERT field trial
185
Table 3), highlighting that it causes a larger range of transfer resistance values.
Figure 5 Joint estimation of leak and petrophysical parameters: the prior and posterior parameter
distributions are shown as violin plots. The true values are marked with orange lines.
3.3. Influence and joint estimation of uncertain (homogeneous) hydraulic
conductivity
Figure 6a shows the estimation of leak parameters under uncertain 𝐾 values.
Estimating leak parameters under 𝐾 uncertainty leads to highly uncertain and
inaccurate leak parameter estimates. Figure 6b shows the estimation of leak
parameters with variance of log𝐾 equal to 2, 3, 5, 7, 10, while assuming the mean
𝐾 values are known exactly and unit correlation lengths. Although some
variations in the estimates are seen, they generally lie close to the true values.
There is no apparent correlation between the leak parameter estimation
performance and the variance of the field. Figure 7 shows the estimates of leak
parameters and effective hydraulic conductivity. The results show good estimates
Synthetic experiments based on the Sellafield ERT field trial
186
of the leak locations, while that for 𝑞 and 𝑡0 is manifested as a narrow envelope.
The posterior uncertainty for 𝐾 remains high and the algorithm underestimates
the effective 𝐾 value. Again, the envelope of mass discharge curves is greatly
reduced, demonstrating a reduction in uncertainty. However, the posterior curves
do not collapse to the true curve, indicating significant uncertainty in the estimates.
Figure 6 (a) The estimation leak parameters under uncertain K values and logK variance = 1.0. The
violin plots show the prior and posterior parameter distributions. The true value is marked with an
orange line. (b) The estimation of leak parameters at variance of log10(K) equal to 2, 3, 5, 7, 10 , while
assuming the mean K values are known exactly and the K field is isotropic and is of unit correlation
length. The violin plots show the posterior parameter distribution, while the true value is marked with
an orange line.
Field application at the Hatfield site
187
Figure 7 (a) Joint estimation of leak parameters and effective hydraulic conductivity. The violin plots
show the prior and posterior parameter distributions. The true value is marked with an orange line. (b)
Prior and posterior tracer mass flux across the pre-defined plane. The true curve is marked in green in
the posterior plot.
4) Field application at the Hatfield site
4.1. Data description
To illustrate the approach in a field setting we use data from a solute injection
experiment at the Hatfield (Yorkshire) site in the UK. At the site, six boreholes were
drilled in 1998 in order to monitor tracer injection, two of which were for transmission
GPR measurements (H-R1 and H-R2), while four were for ERT measurements (H-E1,
H-E2, H-E3, and H-E4). These four ERT boreholes consist of sixteen stainless steel mesh
electrodes equally spaced between 2 and 13 m depth. These boreholes were drilled to
a depth of 12 m and completed with 75 mm PVC casing. Both the ERT and radar
boreholes have a weak sand/cement grout backfilling the annulus. A tracer injection
borehole was also installed (H-I2), located within the centre of the borehole array. The
injection borehole is 3.5 m deep, with a 100 mm diameter slotted section and gravel
pack between 3 and 3.5 m depth.
Field application at the Hatfield site
188
We focus our discussion using the ERT results from the March 2003 tracer
infiltration experiment at Hatfield (Winship et al., 2006). The tracer consisted of 1,200
litres of water, dosed with NaCl to give an electrical conductivity value of 2200 µS
cm-1 (groundwater electrical conductivity at the site was measured as 650 µS cm-1). The
tracer was injected over a period of three days, from 14th March 2003 to 17th March 2003
at a steady rate of approximately 17 litres per hour. A float valve in the injection
borehole was used to control the head in the injection borehole, and hence the flow
rate. Duplicate sets of background measurements of ERT were made on 6th March and
13th March. Tracer flow was monitored by means of a pressure transducer in a storage
tank, which gave a way of calculating the cumulative injection volume over time. The
tracer injection port H-I2 was screened between 3m and 3.5m below ground surface.
The tracer injection was monitored by ERT measurements from four boreholes and
inverted images clearly show the plume migration, as shown in Figure 8 (Winship et
al., 2006). During the tracer test no rainfall was observed at the site. The water table
was observed at approximately 10 m depth.
After removal of outliers, 3108 of the 3172 measurements are kept and 5%
Gaussian data error is assumed in the inversions. Let 𝑡 = 8 be the day where injection
commenced, ERT snapshots for 𝑡 = 7, 10, 15, 21 days are used in the inversion. Table
4 lists the baseline parameters for our simulation, which are largely adopted from
Binley et al. (2002a). The parameters not being estimated are kept constant during
parameter estimation.
Field application at the Hatfield site
189
Figure 8 Setup of the tracer injection test at Hatfield (H-I2 is the injection borehole and H-E1 to H-E4
are ERT boreholes) and the time-lapse resistivity images (iso-surfaces are plotted for 7.5% reduction of
resistivity relative to baseline) obtained from a difference inversion of the ERT data (reproduced from
Winship et al., 2006)
Table 4 Baseline coupled hydrogeophysical model parameters used for the parameter estimation
from the Hatfield field ERT data. *The domain consists of 3 meters of top soil and a uniform main
zone. Only parameters of the main zone are listed below.
PFLOTRAN simulation Value
Total simulation time (days) 41
Model dimensions (m) 30 x 33 x 16
Grid spacing (m) 1 x 1 x 0.5
Permeability (m2) * 4.8225 x 10-13
Porosity * 0.32
Water table depth (m) -12.0
van Genuchten 𝑚 * 0.6
van Genuchten 𝛼 * 3.5 x 10-3
Residual water saturation * 0.04
Recharge (m/day) 1 x 10-4
Leak location (m) (3.0, 4.0,-3.0)
Leak period (day) 8-11
Leak rate (m3/d) 0.408
Background fluid conductivity (S/m) 0.22
Leak fluid conductivity (S/m) 0.065
Mass discharge plane Vertical plane at y = -3 m
E4D simulation Value
Full Model dimensions (m) 500 x 500 x 50
Imaging cell dimensions (m) 10 x 13 x 15
Grid spacing Unstructured
Number of elements 46482
ERT imaging times (day) for inversion 7, 10, 15, 21
Archie’s cementation exponent 1.35
Archie’s saturation exponent 1.35
Field application at the Hatfield site
190
4.2. Parameter estimation
We applied the proposed leak detection framework to the Hatfield field data
and consider two cases (details are listed in Table 5). The first case estimates four leak
parameters and two Archie parameters (𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐, log 𝑞 , 𝑡0,𝑚, 𝑛) . The second case
considers the estimation of a few additional parameters, namely the duration of the
leak event (𝑑𝑡) and the uniform horizontal and vertical hydraulic conductivity (𝐾 and
𝐾𝑧). We consider 𝐾 anisotropy may exist at the site because well logs suggest the
presence of fine layers (Binley et al., 2001). Compared to the earlier synthetic examples,
convergence was much more difficult to achieve. We have used the following
modification to our methods to circumvent this issue: we estimated log 𝑞 instead of 𝑞,
used more realizations, and used a uniform prior instead of a Gaussian one. We
transformed the leak location priors to a uniform grid to aid the interpretation of the
results. We have not considered the estimation of depth of the leak source in any of
our examples because for most leak detection problems, the leak depth is usually
precisely known: for example, base of storage tanks/silos, depth of buried pipelines,
and bottom of landfill lining. Each iteration takes 2.5 hours on average to run on 192
cores. Note that only the forward modelling is parallelized, not the parameter updating.
Results from the base case is reported in Figure 9. Figure 9(a) shows that the
posterior estimates of most parameter pairs form a small cluster. The estimates of 𝑥𝑙𝑜𝑐
and log 𝑞 are close to the true values, while those for 𝑦𝑙𝑜𝑐 and 𝑡0 are slightly above the
true (known) values. The inversion appears to have no sensitivity to 𝑚, while the
estimation of 𝑛 converges to a very small value of about 0.53. Note that in this field test
the true values of 𝑚 and 𝑛 are not known. In the inversion of field data, we would not
necessarily consider the estimates of 𝑚 and 𝑛 representative of actual petrophysical
parameters, but rather they act as hyperparameters to adjust any discrepancy in model
structure. Figure 9(b) shows that the variability of mass discharge curves between
realizations is greatly reduced upon conditioning of ERT data. Specifically, its spread
Field application at the Hatfield site
191
is reduced by two orders of magnitude, highlighting a reduction in site-wide
uncertainty of the plume migration.
Results from the second case are reported in Figure 10. We observe a larger
spread in the parameter space but similar results for the estimation of 𝑚 and 𝑛. 𝑥𝑙𝑜𝑐,
𝑦𝑙𝑜𝑐, and 𝑡0 are slightly overestimated. The inversion appears to have no sensitivity to
𝐾 and 𝐾𝑧. The estimates of log 𝑞 and 𝑑𝑡 centres around the true value, indicating the
inversion algorithm also correctly estimates the total solute loading (𝑞 × 𝑑𝑡) that
enters the flow and transport modelling domain. This underscores that the proposed
data assimilation framework does not suffer from mass balance issues that are
common in inverted resistivity-based approaches.
Table 5 Summary of cases for the Hatfield field example
Figure
Size of
ensemble
(𝑁𝑒)
Parameter(s)
to estimate Prior distribution Comments Final RMSE
Figure 9 64 𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐,
log 𝑞, 𝑡0, 𝑚, 𝑛
Multivariate uncorrelated uniform:
Adjusted uniform grid from 𝑥𝑙𝑜𝑐 0-8m
and 𝑦𝑙𝑜𝑐 =0-10m
log 𝑞= list(min=-2.0, max=1.0),
𝑡0= list(min=0.0, max=20.0),
𝑚 = list(min=0.5, max=2.5),
𝑛 = list(min=0.5, max=2.5)
Base case 223.1615.3
(iter=8,
stabilized
afterwards)
Figure 10 64 𝑥𝑙𝑜𝑐, 𝑦𝑙𝑜𝑐,
log 𝑞, 𝑡0, 𝑑𝑡
𝑚, 𝑛, 𝐾, 𝐾𝑧
Multivariate uncorrelated uniform:
Adjusted uniform grid from 𝑥𝑙𝑜𝑐 0-8m
and 𝑦𝑙𝑜𝑐 =0-10m
log 𝑞= list(min=-2.0, max=1.0),
𝑡0= list(min=0.0, max=20.0),
𝑑𝑡 = list(min=1.0, max=5.0),
𝑚 = list(min=0.5, max=2.5),
𝑛 = list(min=0.5, max=2.5),
𝐾 = list(min=-13.0, max=-9.0),
𝐾𝑧 = list(min=-13.0, max=-9.0)
𝐾, 𝐾𝑧, and
𝑑𝑡 are also
estimated.
310.6613.95
(iter = 2,
stabilized
afterwards)
Field application at the Hatfield site
192
Figure 9 (a) Parameter scatterplots showing pairs of parameter values for the Hatfield example
estimating leak and Archie parameters. The parameter symbols and units are defined in section 3. Grey
squares indicate prior parameter values while black circles in date posterior values. The true leak
parameters used in the field injection experiment is indicated by red triangles. (b) The prior and
posterior mass discharge time series. The sign of mass discharge indicates the direction across the
defined plane.
Field application at the Hatfield site
193
Figure 10 Parameter scatterplots showing pairs of parameter values for the Hatfield example estimating
leak and Archie parameters and hydraulic conductivities. The parameter symbols and units are defined
in Table 4 and section 3. Grey squares indicate prior parameter values while black circles indate
posterior values. The true leak parameters used in the field injection experiment is indicated by red
triangles.
4.3. Global sensitivity analysis using the Morris method
To better understand the sensitivity of ERT data to various parameters in the
coupled hydrogeophysical problem used to analyse the Hatfield dataset, we
performed a global sensitivity analysis using the Morris method (Morris, 1991; Tran et
al., 2016; Wainwright et al., 2014b) that is implemented in the R package sensitivity
(Iooss, 2019). The Morris method returns the elementary effect (EE) of the parameters,
which can be considered as an extension of the local sensitivity method. Since the mean
EE represents the average effect of each parameter over the parameter space, the mean
Field application at the Hatfield site
194
EE can be regarded as a global sensitivity measure. To ignore the effects of the sign,
the mean of absolute EE is usually reported (mean |EE|). In general, for the parameter
ranges considered, parameters with high mean |EE| have a large impact on the data.
Unconditional realizations are generated using the Morris algorithm based on the
parameter ranges specified in Table 6 and the parameter space is sampled uniformly.
We used 25 chains, so for a 13 parameter problem 25 × (13 + 1) = 350 realizations are
generated. We run the forward models using PFLOTRAN-E4D to obtain simulated
ERT response (the settings are the same as those in Table 4, unless otherwise stated).
We set the objective function for calculating the mean |EE| to be the weighted misfit
between the simulated and observed ERT data at Hatfield. The same dataset as in the
previous section is used.
Results from the global sensitivity analysis of the Hatfield experiment shows
that some parameters, especially water table depths and two of the van Genuchten
parameters have the largest effects on the data misfit (Table 6), followed by uniform
permeability, porosity and Archie parameter values. Leak parameters has low mean
|EE|, indicating the difficulty for ERT data to inform their estimation if the others are
not known with confidence. Among them, 𝑥𝑙𝑜𝑐 and 𝑦𝑙𝑜𝑐 have the highest and the
lowest mean|EE|, respectively. Recharge has virtually no effect on the data misfit. The
results show that using ERT data and coupled hydrogeophysical modelling is a
challenging problem. Future work can benefit from better constraining the problem
incorporating additional data sources (e.g. pressure head, concentration, temperature,
saturated and unsaturated hydraulic parameters). Our results agree with that of Tran
et al. (2016), who showed Archie parameters have a higher mean |EE| than van
Genuchten 𝛼. However, they found the mean |EE| of van Genuchten 𝑚 is negligible,
while the largest mean |EE| they found is around 8.0. This highlights the Morris
sensitivity analysis is best considered in a case-to-case basis, as it is affected by the
observed data and the selected parameter ranges. We also report a list of realizations
with low data misfit in Table 6. We observe that none of the realizations have an RMSE
lower than 7.4 and their parameter values vary greatly. It is noteworthy that a “true”
deterministic run (using parameters in Table 4) would give an RMSE of 4.82 (Figure
Field application at the Hatfield site
195
11). The above shows that some solutions to the ERT leak detection problem can be
considered equifinal.
Table 6 Global sensitivity analysis results using the Morris (1991) method on selected parameters on
the Hatfield coupled hydrogeophysical model. The parameter ranges considered and the mean
absolute elementary effect (|EE|) are reported. Parameter value combinations from ten realizations
with the lowest RMSE are also reported.
Parameters
[units] Range Mean|EE| #24 #59 #61 #62 #63 #133 #150 #152 #153 #154
xloc [m] 0.0 – 8.0 7.65 0.0 8.0 2.0 2.0 2.0 6.0 2.0 2.0 2.0 2.0
yloc [m] 0.0 – 10.0 0.18 0.0 10.0 2.5 2.5 2.5 10.0 7.5 7.5 7.5 7.5
q [log10
(m3/d)] -2.0 – 1.0 1.82 -2.00 -1.25 -1.25 -1.25 -1.25 1.00 -1.25 -1.25 -1.25 -1.25
t0 [d] 0.0 – 20.0 1.53 15.00 5.00 5.00 5.00 5.00 15.00 20.00 20.00 20.00 20.00
Archie m [-] 1.0 – 1.5 26.52 1.38 1.00 1.00 1.00 1.00 1.50 1.50 1.50 1.50 1.50
Archie n [-] 0.5 – 2.0 11.68 1.63 0.88 0.88 0.88 0.88 2.00 1.63 1.63 1.63 1.63
water table [m] -14.0 – -9.0 49.39 -14.00 -14.00 -14.00 -14.00 -14.00 -14.00 -12.75 -12.75 -12.75 -12.75
permeability
[log10 (m2)] -15.0 – -12.0 6.85 -15.00 -12.00 -12.00 -14.25 -14.25 -12.00 -12.00 -12.00 -12.00 -12.00
porosity [-] 0.25 – 0.35 12.34 0.25 0.35 0.35 0.35 0.35 0.28 0.35 0.35 0.28 0.28
VG α [Pa-1] 2e-4 – 2e-3 7.50 2.0e-3 2.0e-4 2.0e-4 2.0e-4 1.55e-3 2.0e-4 6.5e-4 6.5e-4 6.5e-4 6.5e-4
VG m [-] 0.4 – 0.8 115.16 0.7 0.7 0.7 0.7 0.7 0.5 0.4 0.7 0.7 0.7
VG Sr [-] 0.01 – 0.2 69.30 0.2 0.01 0.01 0.01 0.01 0.1525 0.01 0.1525 0.1525 0.1525
recharge
[mm/d] 0.0 – 0.001 0.03 0.00 7.5e-4 7.5e-4 7.5e-4 7.5e-4 0.00 1.0e-3 1.0e-3 1.0e-3 2.5e-4
RMSE -- -- 7.54 11.10 11.22 11.15 10.57 9.30 8.25 9.40 7.44 7.42
Figure 11 Transfer resistance scatter plot between the observed and simulated data at Hatfield. The
simulated data uses parameter values listed in Table 4.
Discussion
196
5) Discussion
ERT has been used to detect leaks from nuclear sites for more than two decades.
The conventional approach is to use inversion to obtain smoothed images of resistivity
at different times and to assess whether there is a leak. This approach does not allow
estimation of leak parameters and inversion of large time-lapse ERT datasets can be
computationally demanding. We have presented a data assimilation framework to
estimate leak parameters from ERT data. It evaluates hydrological model proposals
based on the misfits between simulated and observed ERT data and update the model
proposals. The estimated leak parameters are presented as a posterior distribution. It
also outputs plume mass discharge across a plane, which can be used as a metric to
evaluate site-wide uncertainty reduction. These features are not available in existing
methods. Since current methods to estimate mass discharge are based on interpolation
of point measurements, our coupled modelling approach provides an alternative to
quantify mass discharge estimates. Together with point measurements and ERT
imaging, other methods can help establish multiple lines of evidence to better inform
decision making in nuclear site characterisation.
Our synthetic results show that the method allows very good estimation of leak
parameters (e.g. leak rate, loading size, and location). They also show that this
framework can work reasonably well under the influence of uncertain petrophysical
parameters and mean K values, as well as under K heterogeneity with small correlation
lengths. With the rapid growth of autonomous ERT systems to monitor infrastructure,
such as British Geological Survey’s ALERT and PRIME (Huntley et al., 2019) systems,
our approach can provide additional value to ERT data and supplement inverted
resistivity images. Our work also has potential to be applied to other non-point source
leak detection problems such as seepage through embankments, or using a different
geophysical method such as self potential (SP).
We have only examined problems with a few parameters (e.g. leak parameters
and homogeneous Archie and permeability values). All hydrological and
petrophysical parameters that are not being updated are treated as known constants,
which can be strong assumptions on uncertain subsurface properties. Future work
Discussion
197
should strive to relax such assumption and jointly estimate more parameters. The prior
distribution of the uncertain parameters may have an effect on the performance of our
data assimilation approach. Nonetheless, we emphasize that they should be chosen
based on site-specific prior knowledge. In this work, we have considered a relatively
simple problem: a single conservative source with known concentration (thus fluid
conductivity) with a single release episode. With the aid of relevant auxiliary
information, our framework has the potential to be extended to more complex
problems.
The challenges we have encountered when dealing with field data highlights
the need of unbiased and reliable prior information for the proposed method to work
in practice. Equifinality (Beven, 2006; Binley and Beven, 2003) obviously exist in the
leak detection problem since multiple combinations of leak, petrophysical, and
hydraulic parameters can give similar data misfits. Different parameterization, scaling
of parameters, and additional data sources may alleviate the problem. But ultimately,
methods that allow rejection of model proposals may be desirable. Nevertheless, our
method can be considered both a quick and approximate method for quantifying
posterior uncertainty of parameters of interest, as well as a flexible method to perform
regularized inversion without forming the Jacobian (Iglesias, 2016), which can be
advantageous for coupled problems. Our proposed method is best used in well
characterized sites where an abundance of historical data can be used to build prior
models. Alternatively, our method can also be used in controlled tracer injection
experiments to estimate hydraulic, petrophysical and transport parameters.
There exists unique challenges for using raw ERT data in data assimilation. ERT
datasets are usually quite large, with each timeframe containing hundreds to tens of
thousands of data points. The fast collection of ERT data mean that multiple datasets
can be collected daily. However, due to computation constraints, we have only used
data from a few selected days. Also, each ERT quadrupole measurement neither
represent the state response at a point (as in head or concentration data) or the overall
system response (as in hydrocarbon production rates). These challenges do not appear
to impact leak estimation from synthetic results. But their implications warrant further
Conclusions
198
investigation—for example, can we compress raw ERT data for data assimilation since
they may contain significant redundant information?
Frameworks for efficient high-dimensional data assimilation (Ghorbanidehno
et al., 2015; Li et al., 2015, 2016) can be used to jointly estimate a heterogeneous
permeability field. Methods such as level set methods, discrete cosine transform (DCT)
and principal component analysis (PCA) can reduce the number of parameters to
describe a highly heterogeneous field. A recent study has applied ES-MDA in
combination with level set methods (Iglesias and McLaughlin, 2011; Tai and Chan,
2004) to estimate the three-facies heterogeneous permeability field from conservative
tracer test data at the Hanford IFRC site (Song et al., 2019). Future work should explore
their utility in hydrogeophysical data assimilation. Likewise, we have assumed
relatively simple petrophysical relationships in our coupled hydrogeophysical models.
Whether more complex petrophysical models will improve data assimilation results
remains an open question. We also have not examined joint assimilation of ERT data
with head or concentration data, which can be promising for further constraining our
results. In this paper, we have used a relatively small ensemble of highly detailed, fully
coupled hydrogeophysical simulations as the forward model. Our work can benefit
from a recently developed, adaptive multi-fidelity version of ES-MDA (Zheng et al.,
2018), which leverages both the accuracy of highly detailed models and the efficiency
of simplified models within the ES-MDA framework.
6) Conclusions
We propose a data assimilation framework that allows the use of time-lapse
ERT data for solving hydrological parameters in a leak detection problem. It does not
produce any ERT images during inversion; rather, it updates parameters in the
hydrological model to minimize ERT data misfit. The use of an ensemble-based
framework allows straightforward computation of uncertainty estimates. Site-wide
uncertainty reduction can be visualized by comparison of prior and posterior mass
discharge curves. Synthetic and field results demonstrate its utility under a variety of
settings, e.g. when uniform hydrological and Archie parameters are estimated jointly.
Acknowledgement
199
This new framework can be readily extended to solving other complex problems (e.g.
multiple modalities) of interest that is monitored by geophysical data. We have only
used ERT data in our analysis but the framework is highly flexible that it is
straightforward to incorporate multiple data types. Our method complements
electrical resistivity imaging and is particularly applicable to sites where some prior
characterization is performed and uncertainty estimates for the parameters that drive
the underlying processes observed are desired.
7) Acknowledgement
This paper is published with the permission of the Executive Director of the
British Geological Survey (NERC) and Sellafield Ltd. (on behalf of the Nuclear
Decommissioning Authority). James Graham and Nick Atherton provided comments
that helped improve an earlier version of the manuscript. This work is supported by a
Lancaster Environmental Centre PhD studentship and a NDA PhD Bursary; the latter
enabled the first author’s visits to the PNNL to conduct this work. We thank the PNNL
Institutional Computing (PIC) for computing resources. Additional computing
resources was provided by the National Energy Research Scientific Computing Center
(NERSC), a DOE Office of Science User Facility supported by the Office of Science of
the U.S. Department of Energy under Contract No. DE-AC02-05CH11231.
References
Aalstad, K., Westermann, S., Schuler, T.V., Boike, J., Bertino, L., 2018. Ensemble-based
assimilation of fractional snow-covered area satellite retrievals to estimate the
snow distribution at Arctic sites. Cryosphere 12, 247–270.
https://doi.org/10.5194/tc-12-247-2018
Abdulsamad, F., Revil, A., Soueid Ahmed, A., Coperey, A., Karaoulis, M., Nicaise, S.,
Peyras, L., 2019. Induced polarization tomography applied to the detection and
the monitoring of leaks in embankments. Eng. Geol. 254, 89–101.
https://doi.org/10.1016/J.ENGGEO.2019.04.001
Aghasi, A., Mendoza-Sanchez, I., Miller, E.L., Ramsburg, C.A., Abriola, L.M., 2013. A
geometric approach to joint inversion with applications to contaminant source
References
200
zone characterization. Inverse Probl. 29, 115014. https://doi.org/10.1088/0266-
5611/29/11/115014
Audebert, M., Clément, R., Touze-Foltz, N., Günther, T., Moreau, S., Duquennoi, C.,
2014. Time-lapse ERT interpretation methodology for leachate injection
monitoring based on multiple inversions and a clustering strategy (MICS). J.
Appl. Geophys. 111, 320–333. https://doi.org/10.1016/j.jappgeo.2014.09.024
Balbarini, N., Rønde, V., Maurya, P., Fiandaca, G., Møller, I., Erik Klint, K.,
Christiansen, A. V., Binning, P.J., Bjerg, P.L., 2018. Geophysics Based
Contaminant Mass Discharge Quantification Downgradient of a Landfill and a
Former Pharmaceutical Factory. Water Resour. Res.
https://doi.org/10.1029/2017WR021855
Barnett, D.B., Gee, G.W., Sweeney, M.D., Johnson, M.D., Medina, V.F., Khan, F., 2003.
Results of Performance Evaluation Testing of Electrical Leak-Detection Methods
at the Hanford Site Mock Tank – FY 2002-2003. Richland, Washington.
Beven, K., 2006. A manifesto for the equifinality thesis. J. Hydrol. 320, 18–36.
https://doi.org/10.1016/j.jhydrol.2005.07.007
Binley, A., Beven, K., 2003. Vadose zone flow model uncertianty as conditioned on
geophysical data. Groundwater 41, 119–127.
Binley, A., Cassiani, G., Middleton, R., Winship, P., 2002. Vadose zone flow model
parameterisation using cross-borehole radar and resistivity imaging. J. Hydrol.
267, 147–159. https://doi.org/10.1016/S0022-1694(02)00146-4
Binley, A., Hubbard, S.S., Huisman, J.A., Revil, A., Robinson, D.A., Singha, K., Slater,
L.D., 2015. The emergence of hydrogeophsyics for improved understanding of
subsurface processes over multiple scales. Water Resour. Res. 51, 3837–3866.
https://doi.org/10.1002/2015WR017016
Binley, A., Winship, P., Middleton, R., Pokar, M., West, J., 2001. High-resolution
characterization of vadose zone dynamics using cross-borehole radar. Water
Resour. Res. 37, 2639–2652. https://doi.org/10.1029/2000WR000089
Brusseau, M.L., Guo, Z., 2014. Assessing contaminant-removal conditions and plume
persistence through analysis of data from long-term pump-and-treat operations.
J. Contam. Hydrol. 164, 16–24. https://doi.org/10.1016/j.jconhyd.2014.05.004
Cai, Z., Wilson, R.D., Cardiff, M.A., Kitanidis, P.K., 2011. Increasing Confidence in
Mass Discharge Estimates Using Geostatistical Methods. Ground Water 49, 197–
208. https://doi.org/10.1111/j.1745-6584.2010.00709.x
References
201
Camporese, M., Cassiani, G., Deiana, R., Salandin, P., 2011. Assessment of local
hydraulic properties from electrical resistivity tomography monitoring of a
three-dimensional synthetic tracer test experiment. Water Resour. Res. 47, 1–15.
https://doi.org/10.1029/2011WR010528
Camporese, M., Cassiani, G., Deiana, R., Salandin, P., Binley, A., 2015. Coupled and
uncoupled hydrogeophysical inversions using ensemble Kalman filter
assimilation of ERT-monitored tracer test data. Water Resour. Res. 51, 3277–3291.
https://doi.org/10.1002/2014WR016017
Chambers, J.E., Kuras, O., Meldrum, P.I., Ogilvy, R.D., Hollands, J., 2006. Electrical
resistivity tomography applied to geologic, hydrogeologic, and engineering
investigations at a former waste-disposal site. Geophysics 71, B231–B239.
https://doi.org/10.1190/1.2360184
Chen, X., Hammond, G.E., Murray, C.J., Rockhold, M.L., Vermeul, V.R., Zachara, J.M.,
2013. Application of ensemble-based data assimilation techniques for aquifer
characterization using tracer data at Hanford 300 area. Water Resour. Res. 49,
7064–7076. https://doi.org/10.1002/2012WR013285
Chen, Z., Gómez-Hernández, J.J., Xu, T., Zanini, A., 2018. Joint identification of
contaminant source and aquifer geometry in a sandbox experiment with the
restart ensemble Kalman filter. J. Hydrol. 564, 1074–1084.
https://doi.org/10.1016/J.JHYDROL.2018.07.073
Christ, J.A., Ramsburg, C.A., Pennell, K.D., Abriola, L.M., 2010. Predicting DNAPL
mass discharge from pool-dominated source zones. J. Contam. Hydrol.
https://doi.org/10.1016/j.jconhyd.2010.02.005
Christ, J.A., Ramsburg, C.A., Pennell, K.D., Abriola, L.M., 2006. Estimating mass
discharge from dense nonaqueous phase liquid source zones using upscaled
mass transfer coefficients: An evaluation using multiphase numerical
simulations. Water Resour. Res. 42, 1–13. https://doi.org/10.1029/2006WR004886
Crestani, E., Camporese, M., Salandin, P., 2015. Assessment of hydraulic conductivity
distributions through assimilation of travel time data from ERT-monitored tracer
tests. Adv. Water Resour. 84, 23–36.
https://doi.org/10.1016/j.advwatres.2015.07.022
Daily, W., Ramirez, A., Binley, A., 2004. Remote monitoring of leaks in storage tanks
using electrical resistance tomography: Application at the Hanford Site. J.
Environ. Eng. Geophys. 9, 11–24. https://doi.org/10.4133/jeeg9.1.11
References
202
Datta, B., Chakrabarty, D., Dhar, A., 2009. Simultaneous identification of unknown
groundwater pollution sources and estimation of aquifer parameters. J. Hydrol.
376, 48–57. https://doi.org/10.1016/J.JHYDROL.2009.07.014
Emerick, A.A., Reynolds, A.C., 2013. Ensemble smoother with multiple data
assimilation. Comput. Geosci. 55, 3–15.
https://doi.org/10.1016/j.cageo.2012.03.011
Ghorbanidehno, H., Kokkinaki, A., Li, J.Y., Darve, E., Kitanidis, P.K., 2015. Real-time
data assimilation for large-scale systems: The spectral Kalman filter. Adv. Water
Resour. 86, 260–272. https://doi.org/10.1016/j.advwatres.2015.07.017
Hadley, P.W., Newell, C.J., 2012. Groundwater Remediation: The Next 30 Years.
Ground Water 50, 669–678. https://doi.org/10.1111/j.1745-6584.2012.00942.x
Hammond, G.E., Lichtner, P.C., 2010. Field-scale model for the natural attenuation of
uranium at the Hanford 300 Area using high-performance computing. Water
Resour. Res. 46, 1–31. https://doi.org/10.1029/2009WR008819
Hermans, T., Nguyen, F., Caers, J., 2015. Uncertainty in training image-based inversion
of hydraulic head data constrained to ERT data:Workflow and case study. Water
Resour. Res. 51, 5332–5352. https://doi.org/10.1002/2014WR016259
Hermans, T., Oware, E., Caers, J., 2016. Direct prediction of spatially and temporally
varying physical properties from time-lapse electrical resistance data. Water
Resour. Res. 52, 7262–7283. https://doi.org/10.1002/2016WR019126
Hinnell, A.C., Ferr, T.P.A., Vrugt, J.A., Huisman, J.A., Moysey, S., Rings, J., Kowalsky,
M.B., 2010. Improved extraction of hydrologic information from geophysical
data through coupled hydrogeophysical inversion. Water Resour. Res. 46, 1–14.
https://doi.org/10.1029/2008WR007060
Huntley, D., Bobrowsky, P., Hendry, M., Macciotta, R., Elwood, D., Sattler, K., Best, M.,
Chambers, J., Meldrum, P., 2019. Application of multi-dimensional electrical
resistivity tomography datasets to investigate a very slow-moving landslide near
Ashcroft, British Columbia, Canada. Landslides 16, 1033–1042.
https://doi.org/10.1007/s10346-019-01147-1
Iglesias, M., Yang, Y., n.d. A Computational efficient framework for ensemble Kalman
inversion with applications to non-destructive testing and imaging. Prep.
Iglesias, M.A., 2016. A regularizing iterative ensemble Kalman method for PDE-
constrained inverse problems. Inverse Probl. 32, 025002.
https://doi.org/10.1088/0266-5611/32/2/025002
References
203
Iglesias, M.A., McLaughlin, D., 2011. Level-set techniques for facies identification in
reservoir modeling. Inverse Probl. 27, 36. https://doi.org/10.1088/0266-
5611/27/3/035008
Iooss, B., 2019. Package ‘sensitivity.’
Irving, J., Singha, K., 2010. Stochastic inversion of tracer test and electrical geophysical
data to estimate hydraulic conductivities. Water Resour. Res. 46, W11514.
https://doi.org/10.1029/2009WR008340
ITRC, 2010. Use and Measurement of Mass Flux and Mass Discharge, Use and
Measurement of Mass Flux and Mass Discharge. Washington DC.
Johnson, T.C., Hammond, G.E., Chen, X., 2017. PFLOTRAN-E4D: A parallel open
source PFLOTRAN module for simulating time-lapse electrical resistivity data.
Comput. Geosci. 99, 72–80. https://doi.org/10.1016/j.cageo.2016.09.006
Johnson, T.C., Rucker, D.F., Glaser, D.R., 2015a. Near-Surface Geophysics at the
Hanford Nuclear Site, the United States, in: Treatise on Geophysics. Elsevier, pp.
571–595. https://doi.org/10.1016/B978-0-444-53802-4.00205-0
Johnson, T.C., Slater, L.D., Ntarlagiannis, D., Day-Lewis, F.D., Elwaseif, M., 2012.
Monitoring groundwater-surface water interaction using time-series and time-
frequency analysis of transient three-dimensional electrical resistivity changes.
Water Resour. Res. 48, 1–13. https://doi.org/10.1029/2012WR011893
Johnson, T.C., Versteeg, R., Thomle, J., Hammond, G., Chen, X., Zachara, J., 2015b.
Four-dimensional electrical conductivity monitoring of stage-driven river water
intrusion: Accounting for water table effects using a transient mesh boundary
and conditional inversion constraints. Water Resour. Res. 51, 6177–6196.
https://doi.org/10.1002/2014WR016129
Johnson, T.C., Versteeg, R.J., Day-Lewis, F.D., Major, W., Lane, J.W., 2015c. Time-Lapse
Electrical Geophysical Monitoring of Amendment-Based Biostimulation.
Groundwater 53, 920–932. https://doi.org/10.1111/gwat.12291
Johnson, T.C., Versteeg, R.J., Huang, H., Routh, P.S., 2009. Data-domain correlation
approach for joint hydrogeologic inversion of time-lapse hydrogeologic and
geophysical data. Geophysics 74, F127. https://doi.org/10.1190/1.3237087
Johnson, T.C., Versteeg, R.J., Ward, A., Day-Lewis, F.D., Revil, A., 2010. Improved
hydrogeophysical characterization and monitoring through parallel modeling
and inversion of time-domain resistivity and induced-polarization data.
Geophysics 75, WA27. https://doi.org/10.1190/1.3475513
References
204
Johnson, T.C., Wellman, D., 2015. Accurate modelling and inversion of electrical
resistivity data in the presence of metallic infrastructure with known location
and dimension. Geophys. J. Int. 202, 1096–1108.
https://doi.org/10.1093/gji/ggv206
Ju, L., Zhang, J., Meng, L., Wu, L., Zeng, L., 2018. An adaptive Gaussian process-based
iterative ensemble smoother for data assimilation. Adv. Water Resour.
https://doi.org/10.1016/j.advwatres.2018.03.010
Kang, X., Shi, X., Deng, Y., Revil, A., Xu, H., Wu, J., 2018. Coupled hydrogeophysical
inversion of DNAPL source zone architecture and permeability field in a 3D
heterogeneous sandbox by assimilation time-lapse cross-borehole electrical
resistivity data via ensemble Kalman filtering. J. Hydrol. 567, 149–164.
https://doi.org/10.1016/J.JHYDROL.2018.10.019
Kang, X., Shi, X., Revil, A., Cao, Z., Li, L., Lan, T., Wu, J., 2019. Coupled
hydrogeophysical inversion to identify non-Gaussian hydraulic conductivity
field by jointly assimilating geochemical and time-lapse geophysical data. J.
Hydrol. https://doi.org/10.1016/j.jhydrol.2019.124092
Koch, J., Nowak, W., 2016. Identification of contaminant source architectures—A
statistical inversion that emulates multiphase physics in a computationally
practicable manner. Water Resour. Res. 52.
https://doi.org/10.1002/2014WR015716
Köpke, C., Elsheikh, A.H., Irving, J., 2019. Hydrogeophysical Parameter Estimation
Using Iterative Ensemble Smoothing and Approximate Forward Solvers. Front.
Environ. Sci. 7, 34. https://doi.org/10.3389/fenvs.2019.00034
Kowalsky, M.B., Finsterle, S., Peterson, J., Hubbard, S., Rubin, Y., Majer, E., Ward, A.,
Gee, G., 2005. Estimation of field-scale soil hydraulic and dielectric parameters
through joint inversion of GPR and hydrological data. Water Resour. Res. 41, n/a-
n/a. https://doi.org/10.1029/2005WR004237
Kuras, O., Pritchard, J.D., Meldrum, P.I., Chambers, J.E., Wilkinson, P.B., Ogilvy, R.D.,
2009. Monitoring hydraulic processes with Automated time-Lapse Electrical
Resistivity Tomography ( ALERT ). Comptes Rendus Geosci. 351, 868–885.
Kuras, O., Wilkinson, P.B., Meldrum, P.I., Oxby, L.S., Uhlemann, S., Chambers, J.E.,
Binley, A., Graham, J., Smith, N.T., Atherton, N., 2016. Geoelectrical monitoring
of simulated subsurface leakage to support high-hazard nuclear
decommissioning at the Sellafield Site, UK. Sci. Total Environ. 566–567, 350–359.
https://doi.org/10.1016/j.scitotenv.2016.04.212
References
205
Kwong, S., Fowler, L., 2014. Technical Memorandum: MSSS Electrical Resistivity
Tomography (ERT): TRAFFIC modelling of the ERT trial.
LaBrecque, D.J., Ramirez, A.L., Daily, W.D., Binley, A.M., Schima, S.A., 1996. ERT
monitoring on environmental remediation processes. Meas. Sci. Technol. 7, 375–
383. https://doi.org/10.1088/0957-0233/7/3/019
Lan, T., Shi, X., Jiang, B., Sun, Y., Wu, J., 2018. Joint inversion of physical and
geochemical parameters in groundwater models by sequential ensemble-based
optimal design. Stoch. Environ. Res. Risk Assess. 5.
https://doi.org/10.1007/s00477-018-1521-5
Le, D.H., Emerick, A. A., Reynolds, A.C., 2016. An Adaptive Ensemble Smoother With
Multiple Data Assimilation for Assisted History Matching. SPE J. 21, 2195–2207.
https://doi.org/10.2118/173214-PA
Li, J.Y., Kokkinaki, A., Ghorbanidehno, H., Darve, E.F., Kitanidis, P.K., 2015. The
compressed state Kalman filter for nonlinear state estimation: Application to
large-scale reservoir monitoring. Water Resour. Res. 51, 9942–9963.
https://doi.org/10.1002/2015WR017203
Li, L., Puzel, R., Davis, A., 2018. Data assimilation in groundwater modelling: ensemble
Kalman filter versus ensemble smoothers. Hydrol. Process. 32, 2020–2029.
https://doi.org/10.1002/hyp.13127
Li, Y.J., Kokkinaki, A., Darve, E.F., 2016. Smoothing-based compressed state Kalman
filter for joint state-parameter estimation : applications in reservoir
characterization and CO 2 storage monitoring. Water Resour. Res. 53, 7190–7207.
https://doi.org/10.1002/2016WR020168
Manoli, G., Rossi, M., Pasetto, D., Deiana, R., Ferraris, S., Cassiani, G., Putti, M., 2015.
An iterative particle filter approach for coupled hydro-geophysical inversion of
a controlled infiltration experiment. J. Comput. Phys. 283, 37–51.
https://doi.org/10.1016/J.JCP.2014.11.035
Mariethoz, G., Renard, P., Straubhaar, J., 2010. The direct sampling method to perform
multiple-point geostatistical simulations. Water Resour. Res. 46, 1–14.
https://doi.org/10.1029/2008WR007621
Maurya, P.K., Rønde, V.K., Fiandaca, G., Balbarini, N., Auken, E., Bjerg, P.L.,
Christiansen, A.V., 2017. Detailed landfill leachate plume mapping using 2D and
3D electrical resistivity tomography - with correlation to ionic strength measured
in screens. J. Appl. Geophys. 138, 1–8.
https://doi.org/10.1016/j.jappgeo.2017.01.019
References
206
Michalak, A.M., Shlomi, S., 2007. A geostatistical data assimilation approach for
estimating groundwater plume distributions from multiple monitoring events.
pp. 73–88. https://doi.org/10.1029/171GM08
Millard, S.P., 2013. Designing a Sampling Program. EnvStats An R Packag. Environ.
Stat. 25–61. https://doi.org/10.1007/978-1-4614-8456-1
Morris, M.D., 1991. Factorial Sampling Plans for Preliminary Computational
Experiments, Technometrics.
Nowak, W., 2009. Best unbiased ensemble linearization and the quasi-linear Kalman
ensemble generator. Water Resour. Res. 45.
https://doi.org/10.1029/2008WR007328
Oware, E.K., Moysey, S.M.J., Khan, T., 2013. Physically based regularization of
hydrogeophysical inverse problems for improved imaging of process-driven
systems. Water Resour. Res. 49, 6238–6247. https://doi.org/10.1002/wrcr.20462
Pidlisecky, A., Singha, K., Day-Lewis, F.D., 2011. A distribution-based parametrization
for improved tomographic imaging of solute plumes. Geophys. J. Int. 187, 214–
224. https://doi.org/10.1111/j.1365-246X.2011.05131.x
Power, C., Gerhard, J.I., Karaoulis, M., Tsourlos, P., Giannopoulos, A., 2014. Evaluating
four-dimensional time-lapse electrical resistivity tomography for monitoring
DNAPL source zone remediation. J. Contam. Hydrol. 162–163, 27–46.
https://doi.org/10.1016/j.jconhyd.2014.04.004
Ramirez, A., Daily, W., Binley, A., Labrecque, D., Roelant, D., 1996. Detection of leaks
in underground storage tanks using electrical resistance methods. J. Environ.
Eng. Geophys. 1, 189–203. https://doi.org/10.4133/JEEG1.3.189
Ramirez, A.L., Nitao, J.J., Hanley, W.G., Aines, R., Glaser, R.E., Sengupta, S.K., Dyer,
K.M., Hickling, T.L., Daily, W.D., 2005. Stochastic inversion of electrical
resistivity changes using a Markov Chain Monte Carlo approach. J. Geophys. Res.
110, B02101. https://doi.org/10.1029/2004JB003449
Reynolds, A.C., Zafari, M., Li, G., 2006. Iterative forms of the ensemble Kalman filter,
in: Proceedings of 10th European Conference on the Mathematics of Oil
Recovery. Amsterdam, The Netherlands.
Rossi, M., Manoli, G., Pasetto, D., Deiana, R., Ferraris, S., Strobbia, C., Putti, M.,
Cassiani, G., 2015. Coupled inverse modeling of a controlled irrigation
experiment using multiple hydro-geophysical data. Adv. Water Resour. 82, 150–
165. https://doi.org/10.1016/J.ADVWATRES.2015.03.008
References
207
Scholer, M., Irving, J., Looms, M.C., Nielsen, L., Holliger, K., 2012. Bayesian Markov-
chain-Monte-Carlo inversion of time-lapse crosshole GPR data to characterize
the vadose zone at the Arrenaes site, Denmark. (Special Section: Model-data
fusion in the vadose zone.). Vadose Zo. J. 11, 153.
https://doi.org/10.2136/vzj2011.0153
Schöniger, A., Nowak, W., Hendricks Franssen, H.-J., 2012. Parameter estimation by
ensemble Kalman filters with transformed data: Approach and application to
hydraulic tomography. Water Resour. Res. 48, W04502.
https://doi.org/10.1029/2011WR010462
Shlomi, S., Michalak, A.M., 2007. A geostatistical framework for incorporating
transport information in estimating the distribution of a groundwater
contaminant plume. Water Resour. Res. 43.
https://doi.org/10.1029/2006WR005121
Singha, K., Gorelick, S.M., 2006. Effects of spatially variable resolution on field-scale
estimates of tracer concentration from electrical inversions using Archie’s law.
Geophysics 71, G83. https://doi.org/10.1190/1.2194900
Slater, L.D., Ntarlagiannis, D., Day-Lewis, F.D., Mwakanyamale, K., Versteeg, R.J.,
Ward, A., Strickland, C., Johnson, C.D., Lane, J.W., 2010. Use of electrical imaging
and distributed temperature sensing methods to characterize surface water-
groundwater exchange regulating uranium transport at the Hanford 300 Area,
Washington. Water Resour. Res. 46, 1–13. https://doi.org/10.1029/2010WR009110
Song, X., Chen, X., Ye, M., Dai, Z., Hammond, G., Zachara, J.M., 2019. Delineating
Facies Spatial Distribution by Integrating Ensemble Data Assimilationand
Indicator Geostatistics With Level‐Set Transformation. Water Resour. Res.
2018WR023262. https://doi.org/10.1029/2018WR023262
Stordal, A.S., Elsheikh, A.H., 2015. Iterative ensemble smoothers in the annealed
importance sampling framework. Adv. Water Resour. 86, 231–239.
https://doi.org/10.1016/j.advwatres.2015.09.030
Sun, A.Y., 2007. A robust geostatistical approach to contaminant source identification.
Water Resour. Res. https://doi.org/10.1029/2006WR005106
Sun, A.Y., Painter, S.L., Wittmeyer, G.W., 2006. A constrained robust least squares
approach for contaminant release history identification. Water Resour. Res. 42.
https://doi.org/10.1029/2005WR004312
Sun, N.-Z., Sun, A., 2015. Model Calibration and Parameter Estimation. Springer New
York, New York, NY. https://doi.org/10.1007/978-1-4939-2323-6
References
208
Tai, X., Chan, T.F., 2004. A survey on multiple level set methods with applications for
identifying piecewise constant functions. Int. J. Numer. Anal. Model 1, 25–48.
Tran, A.P., Dafflon, B., Hubbard, S.S., Kowalsky, M.B., Long, P., Tokunaga, T.K.,
Williams, K.H., 2016. Quantifying shallow subsurface water and heat dynamics
using coupled hydrological-thermal-geophysical inversion. Hydrol. Earth Syst.
Sci. 20, 3477–3491. https://doi.org/10.5194/hess-20-3477-2016
Troldborg, M., Nowak, W., Lange, I. V., Santos, M.C., Binning, P.J., Bjerg, P.L., 2012.
Application of Bayesian geostatistics for evaluation of mass discharge
uncertainty at contaminated sites. Water Resour. Res. 48, 1–19.
https://doi.org/10.1029/2011WR011785
Troldborg, M., Nowak, W., Tuxen, N., Bjerg, P.L., Helmig, R., Binning, P.J., 2010.
Uncertainty evaluation of mass discharge estimates from a contaminated site
using a fully Bayesian framework. Water Resour. Res. 46, 1–19.
https://doi.org/10.1029/2010WR009227
Truex, M.J., Johnson, T.C., Strickland, C.E., Peterson, J.E., Hubbard, S.S., 2013.
Monitoring Vadose Zone Desiccation with Geophysical Methods. Vadose Zo. J.
12. https://doi.org/10.2136/vzj2012.0147
Truex, M.J., Oostrom, M., Strickland, C.E., Chronister, G.B., Benecke, M.W., Johnson,
C.D., 2012. Field-Scale Assessment of Desiccation Implementation for Deep
Vadose Zone Contaminants. Vadose Zo. J. 11, 0.
https://doi.org/10.2136/vzj2011.0144
Tso, C.-H.M., Kuras, O., Wilkinson, P.B., Uhlemann, S., Chambers, J.E., Meldrum, P.I.,
Graham, J., Sherlock, E.F., Binley, A., 2017. Improved characterisation and
modelling of measurement errors in electrical resistivity tomography (ERT)
surveys. J. Appl. Geophys. 146, 103–119.
https://doi.org/10.1016/J.JAPPGEO.2017.09.009
Wagner, B.J., 1992. Simultaneous parameter estimation and contaminant source
characterization for coupled groundwater flow and contaminant transport
modelling. J. Hydrol. 135, 275–303. https://doi.org/10.1016/0022-1694(92)90092-A
Wainwright, H.M., Finsterle, S., Jung, Y., Zhou, Q., Birkholzer, J.T., 2014. Making sense
of global sensitivity analyses. Comput. Geosci. 65, 84–94.
https://doi.org/10.1016/j.cageo.2013.06.006
Wallin, E.L., Johnson, T.C., Greenwood, W.J., Zachara, J.M., 2013. Imaging high stage
river-water intrusion into a contaminated aquifer along a major river corridor
References
209
using 2-D time-lapse surface electrical resistivity tomography. Water Resour. Res.
49, 1693–1708. https://doi.org/10.1002/wrcr.20119
Ward, W.O.C., Wilkinson, P.B., Chambers, J.E., Nilsson, H., Kuras, O., Bai, L., 2016.
Tracking tracer motion in a 4-D electrical resistivity tomography experiment.
Water Resour. Res. 52, 4078–4094. https://doi.org/10.1002/2015WR017958
Winship, P., Binley, A., Gomez, D., 2006. Flow and transport in the unsaturated
Sherwood Sandstone: characterization using cross-borehole geophysical
methods. Geol. Soc. London, Spec. Publ. 263, 219–231.
https://doi.org/10.1144/GSL.SP.2006.263.01.12
Xu, T., Gómez-Hernández, J.J., 2018. Simultaneous identification of a contaminant
source and hydraulic conductivity via the restart normal-score ensemble Kalman
filter. Adv. Water Resour. 112, 106–123.
https://doi.org/10.1016/j.advwatres.2017.12.011
Xu, T., Gómez-Hernández, J.J., 2016. Joint identification of contaminant source location,
initial release time, and initial solute concentration in an aquifer via ensemble
Kalman filtering. Water Resour. Res. 52, 6587–6595.
https://doi.org/10.1002/2016WR019111
Zeng, L., Shi, L., Zhang, D., Wu, L., 2012. A sparse grid based Bayesian method for
contaminant source identification. Adv. Water Resour. 37, 1–9.
https://doi.org/10.1016/J.ADVWATRES.2011.09.011
Zhang, Y., Liu, N., Oliver, D.S., 2010. Ensemble filter methods with perturbed
observations applied to nonlinear problems. Comput. Geosci. 14, 249–261.
https://doi.org/10.1007/s10596-009-9149-7
Zheng, Q., Zhang, J., Xu, W., Wu, L., Zeng, L., 2018. Adaptive Multi‐Fidelity Data
Assimilation for Nonlinear Subsurface Flow Problems. Water Resour. Res.
2018WR023615. https://doi.org/10.1029/2018WR023615
Zhou, H., Gómez-Hernández, J.J., Li, L., 2014. Inverse methods in hydrogeology:
Evolution and recent trends. Adv. Water Resour. 63, 22–37.
https://doi.org/10.1016/j.advwatres.2013.10.014
Zovi, F., Camporese, M., Hendricks Franssen, H.J., Huisman, J.A., Salandin, P., 2017.
Identification of high-permeability subsurface structures with multiple point
geostatistics and normal score ensemble Kalman filter. J. Hydrol. 548, 208–224.
https://doi.org/10.1016/j.jhydrol.2017.02.056
Paper 3: On the field estimation of moisture content using electrical geophysics—the
impact of petrophysical model uncertainty
210
6. Paper 3: On the field estimation of moisture content using
electrical geophysics—the impact of petrophysical model
uncertainty
Published as Tso, C.-H.M., Kuras, O., Binley, A. (2019): On the field estimation of
moisture content using electrical geophysics‐the impact of petrophysical model
uncertainty. Water Resources Research, 55, 2019, DOI:10.1029/2019WR024964.
Abstract
211
Abstract
The spatiotemporal distribution of pore water in the vadose zone can have a critical
control on many processes in the near‐surface Earth, such as the onset of landslides,
crop yield, groundwater recharge, and runoff generation. Electrical geophysics has
been widely used to monitor the moisture content (𝜃) distribution in the vadose zone
at field sites, and often resistivity (𝜌) or conductivity (𝜎) is converted to moisture
contents through petrophysical relationships (e.g. Archie's law). Though both the
petrophysical relationships (i.e. choices of appropriate model and parameterisation)
and the derived moisture content are known to be subject to uncertainty, they are
commonly treated as exact and error‐free. This study examines the impact of uncertain
petrophysical relationships on the moisture content estimates derived from electrical
geophysics. We show from a collection of data from multiple core samples that
significant variability in the 𝜃(𝜌) relationship can exist. Using rules of error
propagation, we demonstrate the combined effect of inversion and uncertain
petrophysical parameterization on moisture content estimates and derive their
uncertainty bounds. Through investigation of a water injection experiment, we
observe that the petrophysical uncertainty yields a large range of estimated total
moisture volume within the water plume. The estimates of changes in water volume,
however, generally agree within (large) uncertainty bounds. Our results caution
against solely relying on electrical geophysics to estimate moisture content in the field.
The uncertainty propagation approach is transferrable to other field studies of
moisture content estimation.
Key points:
Field evidence demonstrating strong variability of 𝜃(𝜌) relationships at the site
is provided
The proposed methods show the impact of different uncertain 𝜃(𝜌) models on
𝜃 estimates from ERT and their associated uncertainty bounds
Nevertheless, different Archie models give consistent difference in 𝜃 estimates,
though their uncertainty bounds are large
Plain language summary
212
Plain language summary
Maps and images of electrical resistivity have been widely applied to effectively
monitor the wetting or drying of the Earths' near‐surface. But how well can they
quantify such change? How variable are the petrophysical model parameters that
relate resistivity and moisture content? Does uncertainty in such relationships impact
our confidence in moisture content estimates from resistivity imaging? Our analysis of
field samples collected at a UK field site reveals great variability in petrophysical
parameters. Using a uncertainty propagation method, which combines the uncertainty
contributions from both petrophysical parameters and resistivity data errors, we find
that the variable petrophysical parameters can lead to high uncertainty in moisture
content estimates and they appear to be the dominating factor in many cases. These
effects on uncertainty are greater than previously appreciated. The implication is that
realistic uncertainty bounds are needed whenever electrical geophysical methods are
used to quantify the amount of water present underground or its changes over time.
The findings highlight the importance of better characterization of petrophysical
parameters and the need to supplement the interpretation of resistivity‐based moisture
content estimates with other data sources.
1) Introduction
Monitoring the amount of moisture in the Earth's near‐surface is critical in
many applications. For example, the distribution of soil moisture is an important
trigger for landslides (Ray and Jacobs, 2007). The amount of water available for root
water uptake is the most important factor for crop yield (Ahmed et al., 2018). Similarly,
the saturation of the vadose zone governs the rate of groundwater recharge and travel
times of surface contaminants (e.g., nitrate) to an aquifer (Green et al., 2018; Turkeltaub
et al., 2018).
The measurement of moisture content ( 𝜃 ) in the subsurface is not
straightforward. Point sampling can only cover a small number of discrete points in
an investigation area and can be labor‐intensive. These point data may not be
representative of site‐scale variability. In addition, intrusive sampling may disrupt the
Introduction
213
critical processes occurring in the soil (e.g., root growth). Alternative field methods are
needed to improve our ability to measure and monitor moisture content. A
comprehensive review of the different ground‐based methods to determine soil
moisture is given by Jonard et al. (2018).
The well‐established correlation between moisture content and the bulk
resistivity (𝜌) in porous media (Glover, 2015; Lesmes and Friedman, 2005) allows the
use of electrical methods (e.g., electrical resistivity tomography [ERT] and
electromagnetic induction [EMI]) to be applied to study vadose zone processes. They
can be used to derive 2‐D or 3‐D distributed resistivity models over a relatively large
area, and these resistivity models can, in turn, be used for translation to moisture
content via petrophysical relationships. ERT or EMI offers much larger spatial
coverage than point‐based methods without disrupting the Earth materials.
Specifically, ERT is typically performed in transects or between boreholes, while EMI
tends to provide even greater spatial coverage since it is commonly used for mapping.
When applied in time‐lapse mode, they can be a powerful tool to reveal temporal
variations in soil moisture (Robinson et al., 2009).
Over the past two decades, electrical geophysics has been widely used in many
applications in the vadose zone, and increasingly the resistivity images are translated
to obtain quantitative estimates of moisture content. Examples of these applications
include monitoring the onset of landslides (Lehmann et al., 2013; Uhlemann et al., 2017),
hillslope moisture dynamics (Bass et al., 2017; Cassiani et al., 2009; Hübner et al., 2015;
Yamakawa et al., 2012), seasonal changes in soil moisture dynamics (Amidu and
Dunbar, 2007; Binley et al., 2002b), root zone water uptake (Beff et al., 2013; Brillante et
al., 2015; Garré et al., 2011), unfrozen moisture in permafrost (Oldenborger and
LeBlanc, 2015), soil moisture profiles beneath different wheat genotypes (Shanahan et
al., 2015), watershed characterization (Miller et al., 2008), and wetland dynamics
(Chambers et al., 2014b; Scaini et al., 2017; Uhlemann et al., 2016). Previous laboratory
studies have shown that ERT is suitable for characterizing moisture content dynamics
and tracer breakthrough in the unsaturated zone (e.g. Koestel et al., 2008; Wehrer and
Slater, 2015).
Introduction
214
To translate resistivities to moisture content, a petrophysical relationship needs
to be determined. (Note that although the root “petro” implies an application related
to rocks [as in this study], similar physical laws applies to soils as well.) One common
method is to take core samples from the field for laboratory testing (Amidu and
Dunbar, 2007) using well‐established procedures (see Hen-Jones et al., 2017;
Jayawickreme et al., 2008). The samples are often oven dried and re‐wetted, and their
resistivities are then repeatedly measured as their saturation changes. Although
hysteresis has been reported in the wetting‐drying behavior of samples, laboratory
testing is usually only applied to a single drying or wetting regime. Another method
is to calibrate field‐based inverted resistivity from ERT with in situ measurements of
soil moisture, for example, using time domain reflectometry (TDR) probes. Several
studies have compared moisture content estimates from TDR and ERT (Brunet et al.,
2010), and in recent years it has become increasingly popular to use such field‐derived
petrophysical relationships. The local TDR‐derived moisture content is taken as error‐
free, and this is typically used to calibrate against inverted resistivities using Archie's,
Waxman–Smits (Cassiani et al., 2009; Garré et al., 2013; Lehmann et al., 2013; Michot et
al., 2003), or data‐driven models (Brillante et al., 2014). More recently, calibration
methods have been developed for apparent electrical conductivity from EMI against
TDR‐derived moisture content (Robinet et al., 2018). The repeated EMI‐moisture
content monitoring study of Martini et al. (2017)shows that this is not as
straightforward as the relationship between electrical conductivity and moisture
content can change with time. Whalley et al. (2017) compared the change in electrical
conductivity from EMI and ERT with changes in water content from neutron probe
measurements. The third (and perhaps most common) option is to simply use
literature values for petrophysical parameters (e.g., Friedman, 2005). Regardless of the
method for the assignment of petrophysical relationships, errors will be present in
some form. Laboratory measurements assume the observed relationship and errors
from small samples taken at a few locations can be applied to the entire resistivity
model. Field‐based petrophysical relationships, on the other hand, assume the inverted
resistivity model having insignificant and uncorrelated errors so that they can be used
Introduction
215
to calibrate against in situ soil moisture data. In other words, the resistivity model
uncertainty is implicitly counted twice.
The uncertainty of the moisture content estimates from electrical geophysics
not only stems from the uncertainty in the resistivity model, but it also propagates
through from any constitutive relationships linking geophysical and hydrological
properties, and yet these relationships are frequently assumed to be precise and error‐
free (Binley et al., 2015), in part due to the time and effort required to measure
petrophysical parameters in the lab. In fact, they are known to be uncertain due to the
competing properties of the pore fluids, pore geometry, and pore surface area on
resistivity measurements (Weller et al., 2013). Petrophysical model uncertainty is also
one of the primary factors limiting the utility of coupled inversion approaches (i.e.,
joint estimation of geophysical and hydraulic properties; Singha et al., 2015). While
some stochastic modeling approaches (e.g. Hermans et al., 2015; Hinnell et al., 2010;
Wiese et al., 2018) allow some modifications so that petrophysical model uncertainty
can be accounted for, resolving issues caused by such uncertainty remains an area of
research. Recent coupled inversion approaches allow the option to jointly estimate
petrophysical parameters. Kuhl et al. (2018) devised a coupled inversion approach to
jointly estimate soil hydraulic parameters, petrophysical parameters, and root
parameters simultaneously. Such methods are promising, but there are concerns over
the non‐uniqueness in the inverse problem formulation and that the petrophysical
parameters obtained may merely be “effective” ones. In summary, research is needed
to investigate the extent of the impact on moisture content estimates due to uncertain
petrophysical relationships.
The oil and gas industry, from where many of the foundational petrophysical
relationships used in hydrogeophysics are borrowed, or originate, has been aware of
the potential impact of petrophysical uncertainty. For example, Glover (2017)
highlighted that various sources of uncertainties in Archie parameters can lead to 20–
40% error in hydrocarbon saturation. For instance, even an uncertainty of 0.01 in a
saturation exponent of 2 (i.e., 0.5% or 2 ± 0.01) would result in an error in global oil
reserves of about USD ±254.36 billion based on figures in December 2015. While it is
Introduction
216
difficult to put a monetary value on many near‐surface applications, the above
calculation underscores the highly sensitive nature of petrophysical parameters, and
one should anticipate a similar scale of error in soil water content estimation from
electrical hydrogeophysics.
It is not until recently that the issues associated with petrophysical uncertainty
have been investigated. The pioneering work of Brunetti et al. (2018) considered the
effect of petrophysical uncertainty on using ground penetrating radar (GPR) data for
Bayesian hydrological model selection. There has also been some study on the
parameter uncertainty of petrophysical models. For instance, Laloy et al. (2011) tested
five “pedo‐electrical” models for the reproduction of electrical resistivity (determined
by ERT) in a silt loam soil sample across a range of moisture and bulk density values.
They were inverted within a Bayesian framework, thereby identifying not only the
optimal parameter set but also the parameter uncertainty and its effect on model
prediction. However, to date, there has not been any study on how the uncertainty of
petrophysical relationships affects the quantitative estimation of soil water in the
vadose zone using electrical geophysics. The findings on this question are relevant to
many applications mentioned above.
In this work, we present a first attempt to investigate the extent to which
moisture content estimates are affected by uncertainty in petrophysical models. Our
aims are to understand the likely variability in petrophysical models and to develop a
method for petrophysical uncertainty propagation, which can be used to explore
contributions to uncertainty in the estimation of soil moisture. We review time‐lapse
ERT monitoring data of a controlled infiltration experiment and the rock core data
collected in the same formation. We test the two types of petrophysical models on the
core data and apply it to the inverted resistivity model, while keeping track of the
uncertainty propagation quantitatively. The methods and data used in this work are
detailed in section 2. We report results from our analysis in section 3. Finally, we
discuss our findings in section 4 and provide our conclusions in section 5.
Materials and methods
217
2) Materials and methods
Our study focuses on data from earlier comprehensive field and laboratory
investigations, at Hatfield (near Doncaster, South Yorkshire, UK) and Eggborough
(near Selby, North Yorkshire, UK). Two field sites, 17 km apart from each other, were
instrumented to study recharge processes to a Sherwood Sandstone aquifer. Tracer
injection experiments, monitored by both ERT and GPR, were performed at both sites.
At Eggborough, ERT and GPR surveys were conducted in 1999 (Binley et al., 2002a;
Cassiani and Binley, 2005), and the data were used to study the utility of joint inversion
of ERT and GPR data (Bouchedda et al., 2012; Linde et al., 2006) and the influence of
prior information on vadose zone parameters estimation in stochastic inversion
(Scholer et al., 2011). Similarly, both ERT and GPR surveys were conducted during
tracer injection at Hatfield, and they have been used in a series of studies to improve
the monitorability and predictability of vadose zone processes using geophysical
measurements (Binley et al., 2004, 2002a, 2002b, 2001; Binley and Beven, 2003). Two
radar and four ERT boreholes were drilled around an injector to monitor tracer
injection. Each ERT borehole consists of 16 stainless steel mesh electrodes equally
spaced at 0.733 m between 2 and 13 m depth. The borehole electrodes were
supplemented with eight surface electrodes. Two cored boreholes were drilled close to
the tracer injection area to obtain a depth profile of grain size distribution. Note that
the top 2 meters is topsoil while its underlying material is weakly cemented sandstone.
A similar borehole ERT and GPR setup was applied for the monitoring experiment at
the Arreneas infiltration plant in Denmark (Haarder et al., 2012; Looms et al., 2008).
In this study, we fitted the Archie relationships for the cores collected at
Eggborough and used them as realizations of petrophysical models. We then
simulated the ERT response of a water injection experiment, assuming a baseline
petrophysical relationship. We then inverted the ERT response and use each of the
realizations of petrophysical models to estimate moisture content with uncertainty
bounds, which we compared against the simulated value. We summarize the
workflow of our approach in Figure 1.
Materials and methods
218
Figure 1 Moisture content (𝜽) estimation and petrophysical uncertainty propagation workflow used in
this study. Rectangles indicate model inputs or data, while ovals represent modeling or analysis steps.
We obtained synthetic ERT and 𝜽 data using PFLOTRAN‐E4D. Then we inverted the ERT data and
used the Eggborough cores as different petrophysical models. They were passed through the moisture
content estimation and uncertainty estimation framework to obtain ERT‐estimated 𝜽, which were
compared against the 𝜽 data. ERT = electrical resistivity tomography.
2.1. Eggborough Core Samples
Core samples collected at Eggborough were used to measure the spectral
induced polarization responses at various saturations (Binley et al., 2005), and they are
compared with various physical and hydraulic properties (Supporting Information
Table S2). They found a strong correlation between mean relaxation time and hydraulic
conductivity and showed that the former is affected by saturation. Binley et al. (2005)
did not include the data showing the direct current (DC) resistivity and hydraulic
properties were not published. Also, they focused their analysis on only three of the
samples extracted. In this work, we examine the DC resistivity–saturation behavior of
all the samples to understand its variability and the impact of such variability on
estimating moisture content from ERT.
The grain size distribution of the Eggborough cores and blocks are plotted as
percentiles (Figure 2a). Also, the percentages of sand, silt, and clay at Eggborough are
plotted as depth profiles (Figure 2b). Note that the cores are not repacked sample but
instead they are weakly cemented core plugs. In this work, we use the Eggborough
data to obtain petrophysical relationships for predicting moisture content in a water
injection simulation.
Materials and methods
219
Figure 2 (a) Cumulative density functions of grain size distribution of Eggborough cores and blocks.
The legend shows the core or block ID. (b) Depth profiles of sand (blue), silt (red), and clay (yellow)
percentages for Eggborough cores.
2.2. Water Injection Simulation
The March 2003 tracer infiltration experiment at Hatfield (Binley, 2003;
Winship et al., 2006) used a tracer that consisted of 1,200 L (or 1.2 m3) of water, dosed
with NaCl to give an 𝜎𝑓 of 2,200 μS/cm (groundwater 𝜎𝑓 was 650 μS/cm). The tracer
was injected over a period of 3 days, from 14 March 2003 to 17 March 2003 at a steady
rate of 17 L/hr. The tracer injection port was screened between 3 and 3.5 m below
ground surface. The water table was at 10 m below ground surface. The layout of the
electrodes is shown in Figure 5.
Since our focus here is the change in moisture content, we numerically repeat
the Hatfield 2003 injection experiment with groundwater instead of a conductive tracer.
We used the parallel coupled hydrogeophysics code PFLOTRAN‐E4D (Johnson et al.,
2017) to simulate the flow and transport of the water injection and to obtain the
corresponding ERT response. PFLOTRAN (Hammond et al., 2014) is a subsurface flow
and reactive transport code, and we use the Richards model to simulate variably
saturated flow. E4D (Johnson et al., 2010) is a 3‐D modeling and inversion code
designed for subsurface imaging and monitoring using static and time‐lapse 3‐D
electrical resistivity or spectral induced polarization data, which we use here as a
Materials and methods
220
forward ERT simulator. The PFLOTRAN grid consists of 129,600 cells that are 0.25 to
1 m wide and 0.5 m thick. The E4D mesh is an unstructured tetrahedral mesh generated
by tetgen (Si, 2015). The resultant mesh comprises 8,124 nodes and 46,842 elements.
PFLOTRAN‐E4D interpolates and maps the PFLOTRAN outputs to electrical
resistivity on the E4D mesh given element‐wise petrophysical transform. ERT
snapshots are taken on Days 7, 9, 10, 15, 18, 21, 27, and 41. We assume a 2%
measurement error in each of the 3,108 measurements taken in each frame. An
additional 2.5% is added to the data errors in the inversions to account for forward
modeling errors. The hydraulic conductivity field is assumed uniform and uses the
values reported in Binley et al. (2002a). The parameters used in the simulation can be
found in Table 1. The assumed petrophysical parameters are also plotted in Figure 4.
Parameters Value Parameters Value
Initial water saturation 0.375 Water fluid conductivity 650 μS/cm
Injector depth interval 3-3.5 m Assumed 𝑛 1.35
Water injection Rate 0.408 m3/d Assumed ρs (at 650 μS/cm) 44 Ω m
Injection period Day 8-11 Assumed ERT data errors 4.5%
Hydraulic conductivity 0.4 m/d van Genuchten 𝛼 10 m−1
Porosity 0.32 van Genuchten 𝑛 2.5
Table 1 Parameters used for the water injection experiment.
2.3. Petrophysical models
2.3.1. Archie’s Law
Assuming a minimal contribution from electrical conductivity on the grain
surface, Archie's law relates bulk electrical resistivity 𝜌 (1/conductivity) to fluid
saturation 𝑆. It is given by
𝜌 = 𝜎𝑓−1𝜙−𝑚𝑆−𝑛 (1)
where 𝑚 is the cementation factor, 𝜎𝑓 is the fluid conductivity, 𝜙 is the porosity, and n
is the saturation exponent. Assuming constant material and fluid properties (e.g., 𝑚, 𝑛,
and 𝜎𝑓), Archie's law can be re‐written in terms of the electrical resistivity at saturation
(i.e., 𝑆 = 1), which is given by
Materials and methods
221
𝑆 = (𝜌𝑠
𝜌)
1
𝑛 (2)
where 𝜌𝑠 = 𝜎𝑓−1𝜙−𝑚 . To obtain best‐fit estimates of Archie parameters, a straight line
is fitted for log10(𝑆) and log10(𝜌𝑠) using the least‐squares criterion. The fitting routine
returns the covariance structure of the model estimates, which can be used to
determine the 68% confidence interval (1 standard deviation) of the model estimates.
Note that 𝜌𝑠 corresponds to a particular 𝜎𝑓 . Therefore, it needs to be scaled when
applied to a different 𝜎𝑓 using equation 1. We note that constant fluid conductivity
may not be appropriate in a range of environments (e.g., Altdorff et al., 2017). Because
the clay content in the cores is low, the results from fitting the Waxman–Smits model
are not reported. Note that saturation and moisture content 𝜃 are related by 𝑆 = 𝜃/𝜙.
The total amount of moisture 𝑉𝑤 within a volume 𝑉 is given by 𝜙𝑉𝑆.
The fractional change of 𝜃, or equivalently that of 𝑆, is given by
𝜃𝑡
𝜃0= (
𝜌𝑡
𝜌0
𝜎𝑓,𝑡
𝜎𝑓,0)
−1
𝑛 (3)
where the subscirpts 𝑡 and 0 represent the variable at time 𝑡 and at baseline.
2.4. ERT modelling and inversion
We use the code R3t version 1.8
(www.es.lancs.ac.uk/people/amb/Freeware/R3t/R3t.htm) for ERT inversion. To obtain
the resistivity variation, we seek to find a model solution that minimizes the following
objective function:
Φ = Φ𝑑 +Φ𝑚 = (𝑑 − 𝐹(𝑚))𝑇𝑊𝑑𝑇𝑊𝑑(𝑑 − 𝐹(𝑚)) + 𝛼𝑚
𝑇𝑅𝑚 (4)
where 𝑑 is the data (e.g., measured apparent resistivities), 𝐹(𝑚) is the set of simulated
data using the forward model and estimated parameters 𝑚. 𝑊𝑑 is a data weight matrix,
which, if we consider the case of uncorrelated measurement error and ignore forward
model errors, is a diagonal matrix with entries equal to the reciprocal of the errors of
each measurement. Forward modeling errors are also added to the diagonal of 𝑊𝑑. 𝛼
Materials and methods
222
is the scalar regularization factor, while 𝑅 is a roughness matrix that describes the
spatial connectedness of the parameter cell values. 𝛼 is selected via a line search, and
isotropic smoothing is applied.
Using a Gauss–Newton procedure, the above is solved iteratively using the following
solution:
(𝐽𝑇𝑊𝑑𝑇𝑊𝑑𝐽 + 𝛼𝑊𝑚
𝑇𝑊𝑚)∆𝑚 = 𝐽𝑇𝑊𝑑(𝑑 − 𝐹(𝑚)) − 𝛼𝑅𝑚𝑘 (5)
𝑚𝑘+1 = 𝑚𝑘 + ∆𝑚
where 𝐽 is the Jacobian (or sensitivity) matrix, given by 𝐽𝑖,𝑗 = 𝜕𝑑𝑖 𝜕𝑚𝑗⁄ ; 𝑚𝑘 is the
parameter set at iteration 𝑘; and ∆𝑚 is the parameter update at iteration 𝑘. For the DC
resistivity case, the inverse problem is typically parameterized using log-transformed
resistivities.
For analysis of time-lapse ERT, we follow the difference inversion approach
(Labrecque and Yang, 2001) to invert on the change in ERT data. Its model penalty
function seeks to minimize model variation relative to a reference mode 𝑚𝑟𝑒𝑓:
Φ𝑚 = 𝛼(𝑚 −𝑚𝑟𝑒𝑓)𝑇𝑅(𝑚 −𝑚𝑟𝑒𝑓) (6)
Again, using a Gauss-Newton procedure, the objective function can be solved
iteratively by:
(𝐽𝑇𝑊𝑚𝑇𝑊𝑚𝐽 + 𝛼𝑅)∆𝑚 = 𝐽𝑇𝑊𝑑 ([(𝑑 − 𝑑𝑟𝑒𝑓) − (𝐹(𝑚) − 𝐹(𝑚𝑟𝑒𝑓))]) − 𝛼𝑅(𝑚 −𝑚𝑟𝑒𝑓)
𝑚𝑘+1 = 𝑚𝑘 + ∆𝑚 (7)
where 𝑑𝑟𝑒𝑓 is the baseline data vector. This approach, which has been proven to be
effective in removing the effect of systematic errors (e.g., artifacts), has been applied to
numerous time‐lapse imaging studies (Doetsch et al., 2012b; LaBrecque et al., 2004).
Note that the same mesh is used for both ERT forward modeling and inversion.
2.5. Uncertainty propagation and moisture content estimation
After inverting the electrical resistivity models, we can obtain the
corresponding element‐wise moisture content using the petrophysical relationships.
Materials and methods
223
The quantity of water within a certain volume is given by the spatial integral of the
moisture content within the volume.
Rules of analytical uncertainty propagation (Chen and Fang, 1986; Taylor, 1982)
were followed to propagate petrophysical uncertainty to moisture content estimates at
each element. The uncertainty of saturation estimated from Archie's law is given by
the following equation (see Appendix A for details):
𝜎𝑆2 = (
𝜕𝑆
𝜕𝜌)2𝜎𝜌2 + (
𝜕𝑆
𝜕𝜌𝑠)2𝜎𝜌𝑠2 + (
𝜕𝑆
𝜕𝑛)2𝜎𝑛2 (8)
where 𝜎2 is the variance of parameters. 𝜎𝜌𝑠2 and 𝜎𝑛
2 are determined by the parameter
fitting procedures. 𝜎𝜌2 are determined by running Monte Carlo simulations of ERT
inversion (Aster et al., 2005; Tso et al., 2017). This procedure, in essence, samples the
measurement errors based on the prescribed error levels and obtains a distribution of
inverted resistivity at each cell due to the perturbed measurements. The first term in
the above equation can be viewed as the variance contribution from the variance of
ERT inversion, while the other terms are the contributions from the uncertainty in the
petrophysical fits. When evaluating the difference in saturation between two survey
times, it is important to take account of the fact that their uncertainties may be
correlated. Therefore, the variance of the difference in saturation 𝛥𝑆 is given by
𝜎∆𝑆 = √𝜎𝑆2 + 𝜎𝑆0
2 − 2cov(𝑆, 𝑆0) (9)
where 𝑆0 is saturation at baseline and cov(𝑆, 𝑆0) is approximated by all the 𝑆 values in
the model domain at the two times. The variance of saturation can be converted to that
of the total amount of water (𝑉𝑤) within a volume by
𝜎𝑉𝑊2 = (
𝜕𝑉𝑊
𝜕𝜙)2𝜎𝜙2 + (
𝜕𝑉𝑊
𝜕𝑆)2𝜎𝑆2 = (𝑉𝑆)2𝜎𝜙
2 + (𝑉𝜙)2𝜎𝑆2 (10)
If porosity 𝜙 is assumed to be known and constant, the first term is dropped. For a
finite element domain consisting of many elements, the total variance is simply the
sum of variances of all the elements.
Results
224
3) Results
3.1. Fitting Archie models
Figure 3 shows the water saturation–electrical resistivity relationship of 12 of the
Eggborough cores and blocks. Note that some sample exhibits rather large scatter, and
in a few occasions, the resistivity shows a decrease with decreasing saturation. Archie's
law is fitted on the data. The best‐fit line and the corresponding ±1 standard deviation
envelope are also plotted. Both 𝜌𝑠 (27.45–64.35 Ω m ) and n (0.513–2.174) show
significant variability. As observed in Table S1, the variability in Archie parameters
does not tend to correlate with texture‐related properties. In most previous studies
literature‐based estimates of Archie parameters are adopted, and where laboratory
analysis is carried out, only a few samples are used. The significant variability (within
the same unit) and lack of correlation with other properties presented here illustrate
the challenge of constraining Archie parameters in the field. Our data show two
distinct groups of clay contents (∼2% and ∼3.5%), and the corresponding Archie
parameters show slightly different ranges. Figure 3 also shows the Archie's parameter
estimation of all Eggborough cores and blocks. The predictions using the best estimate
of the parameters are shown in solid lines, while the 68% (i.e., ±1 standard deviation)
confidence intervals are shown in dashed lines. It shows that when fitting all of the
cores and blocks together, the resultant standard deviation is low, leaving some data
points outside the ±1 standard deviation envelope. We have also included the fit for
Hatfield cores reported in Binley et al. (2002b) and summarize all the Archie models in
Figure 4. Further details, including hydraulic and surface area measurements, of the
Eggborough cores and blocks can be found in Table S2.
Results
225
Figure 3 Archie's parameter estimation of individual Eggborough cores and blocks. The predictions
using the best estimate of the parameters are shown in solid lines, while the 68% (i.e., ±1 standard
deviation) confidence intervals are shown in dashed lines. Note that the measurements are made at
𝝈𝒇 = 𝟏𝟎𝟎𝟎 𝛍𝐒 𝐜𝐦−𝟏. Note that 𝝆, which is the dependent variable, is shown on the x‐axis.
Figure 4 Summary of Archie model fits for the Eggborough/Hatfield cores and blocks. Note that
values correspond to 𝝈𝒇 = 𝟏𝟎𝟎𝟎 𝛍𝐒 𝐜𝐦−𝟏. The point label “synthetic” is the “true” solution
considered in the synthetic study in section 3.2.
Results
226
3.2. Moisture content estimation for the water injection simulation
The time‐lapse ERT monitoring data during the water injection simulation was
inverted using a difference inversion as described above. The iso‐surfaces in Figure S1
show a volume that has 5.5% reduction of resistivity relative to baseline (Day 7). The
inversion results capture the geometry and the swell‐shrink dynamics of the plume
very well. The plume expanded gradually once the injection commenced and then
migrated downward within a few days after the injection finished.
Our subsequent results focus on an ERT snapshot 10 days after the injection
(Day 18). Figures 5a and 5b show the resultant mean and standard deviation of
electrical resistivities obtained from Monte Carlo runs of ERT inversion. Since we have
assumed uniform initial saturation, the variation of resistivity is within the same order
of magnitude. The center region of the ERT array shows reduced resistivity due to
injection. The standard deviation is higher around the electrodes and is lower in the
center region because the resolution of ERT decreases away from electrodes.
Conceptually, however, the uncertainty in the center region through which the water
plume evolves should be higher. This issue is not addressed in this study. Based on the
Monte Carlo inversion results, Figure 5c shows the volume extracted from the ERT
inversion domain where there is at least a 5.5% reduction in resistivity on Day 18
relative to the pre‐injection baseline (Day 7). Such a threshold is used so that the effects
of inversion artifacts are minimized. The size of this volume is 79.97 m3. The total
amount of water in this volume at Days 7 and 18 are 9.65 and 10.68 m3, respectively.
The resistivities on the nodes of the extracted volume were converted to saturation
using the different petrophysical relationships (i.e., Archie model fits) discussed above,
while a Monte Carlo experiment was run to estimate the uncertainty in the inverted
resistivities.
Results
227
Figure 5 (a) Mean (log10) and (b) standard deviation (linear) of electrical resistivity for Day 18 obtained
from Monte Carlo runs of electrical resistivity tomography inversion. (c) Extracted volume where there
was a 5.5% reduction of resistivity relative to baseline on Day 18. The purple cubes are electrode
locations.
For each of the petrophysical models, we then integrate the moisture contents
over the extracted volume to estimate the total water volume (𝑉𝑤) in it. At the same
time, we derive error bars for the total water volume estimates using equations 8 and
9. Figure 6a shows the mean and uncertainty bounds for the amount of water within
the extracted volume, assuming a constant porosity of 0.32. For Day 18 (post‐injection),
best estimates of total water volume among Archie models lie between 8.70 m3
(Binley02) and 16.74 m3 (VEC15‐5), except for VEG2R1 and VEC18‐1 that lie at 2.51 and
3.88 m3, respectively. The size of the error bars varies between ±0.68 m3 (VEG2R1) and
±2.28 m3 (VEG15‐8), or between 9.59% (VEC18‐2) and 27.01% (VEG2R1), depending on
the Archie parameters estimates and their uncertainties. We observe similar results for
Day 7 (pre‐injection), yet we note that while the size of the error bars generally
increases from Day 7 to Day 18, the increase ranges from 0.19 m3 (HEC15‐1) to 0.72 m3
(“all”).
Results
228
Figure 6 (a) Total water volume within the extracted volume (with uncertainty bounds) using the
different petrophysical models. The uncertainty bounds correspond to ±1 standard deviation. The
vertical lines show the true total water volume. (b) The corresponding changes in the amount of
moisture within the extracted volume relative to baseline. The vertical lines show the true change in
total water volume. (c) The contribution of different variables to the variance of total moisture of each
petrophysical models. (d) Additional variance (i.e., uncertainty) caused by uncertain porosity values
(0.32±0.032). The contribution from uncertain porosity is significant in most cases, especially when the
variance in saturation is low.
Figure 6b shows the change in total water volume on Day 18 relative to baseline.
The mean change is the difference between the total water volume at the two times.
Using equation 10, the error bars shown here have accounted for potential correlation
between total water volume estimates between the two times. As a result, when fluid
conductivity is assumed constant, the uncertainty bounds for the change in total
moisture would lie between one and two times of that of the total moisture. The Archie
models estimate an increase in mean change in total water volume of 0.46 m3 (VEG2R1)
to 1.08 m3 (VEG2R2). They are more consistent than the estimates of the absolute total
water volume. Note that the total injection volume was 1.224 m3 , meaning all the
models have underestimated the addition of water due to injection. The uncertainty
Results
229
bounds in Figure 6b are generally large, ranging from ±0.71 m3 (VEG2R1) to ±2.96 m3
(VEC15‐8), or 154% (VEG2R1) to 350% (HEC15‐1) of the mean value. This shows that
even though the mean estimates for the change in total water volume using Archie
models is consistent, they are nevertheless highly uncertain.
The size of the error bars in Figure 6a is determined by a combination of the
uncertainty of the petrophysical parameters (𝜌𝑠 and 𝑛 ) and that of the inverted
resistivities 𝜌. Based on equations 8 and 9, the variance of the total moisture estimates
is the summation of the squared product of the partial derivative and standard
deviation of the individual terms. We plot the terms as stacked bars for Day 18 (post‐
injection) in Figure 6c to show their contribution to the total variance. The square root
of the total height of the bars equals the size of the error bars in Figure 6a. The
contribution from inverted resistivities 𝜌 is below 2 (m3)2 for all the Archie models.
For the Archie models with variance smaller than 2 (m3)2, inverted resistivities can be
an important source of errors; otherwise, the effects of uncertain petrophysical
parameters dominate. Our results indicate that for the Archie models, 𝑛 plays a more
important role than 𝜌𝑠, with the exception of Binley02, which shows very low n error.
𝑛 contributes 3.88% (VEG2R1) to 69.25% (HEC15‐1) of the total variance, while 𝜌𝑠
contributes 2.55% (VEG2R1) to 36.71% (VEC16‐3) of the total variance.
So far we have assumed the porosity has a constant value of 0.32. Additional
uncertainty is introduced if it is treated as uncertain. We consider the case where
porosity is assumed to be 0.32 ± 0.032. In Figure 6d, the height of the blue bars is the
total height of the bars in Figure 6c. The height of the yellow bars shows the additional
variance due to the uncertain porosity value, which ranges from 0.0631 (m3)2
(VEG2R1) to 2.8026 (m3)2 (VEC15‐5). Percentage‐wise, the uncertain porosity values
lead to an increase in variance ranging from 13.7% (VEG2R1) to 108% (VEC18‐2).
We have examined in Figure 6b the change in total moisture within the
extracted volume. We examine in Figure 7 the change in volume of water within each
finite element cell of the extracted volume. Figure 7a shows the estimated change in
the volume of water (𝑉𝑤) in four selected cells. It is observed that while the true change
spans from 0 to 0.18 m3, the estimates for Archie models stay within the 0 to 0.05 m3
Results
230
range. Figure 7b shows the scatter plots for the ERT‐estimated Vw using the 15 Archie
models. For all of them, the fit at individual cells is unsatisfactory. Conversely, in
Figure 6b the changes in total moisture within the extracted volume are fairly
consistent across the petrophysical models, and they agree with the true value. We
observe that within the extracted volume (the threshold was change in inverted
resistivity greater than 5.5%), 101 of 219 cells show change in saturation of less than
0.01. This indicates the true water plume is much narrower than estimated by ERT
inversion and highlights the detection limit of ERT, particularly in the context of
smoothness‐constrained inversion used here. The smoothing effect of the ERT
inversion, however, roughly preserves mass balance in this case.
Figure 7 (a) Electrical resistivity tomography estimated changes in volume of water in four selected
cells. The vertical lines indicate the true change. (b) Scatter plots showing the fit for change in volume
of water at individual cells using the 15 Archie models. The red dashed line in each plot is the best‐fit
line of the scatter points.
Discussion and implications for future research
231
4) Discussion and implications for future research
4.1. Fitting petrophysical models
Most previous studies have either fitted petrophysical models for up to a few
cores or used petrophysical parameters based on literature values without assuming
any errors or uncertainty. Our results from cores collected at a relatively uniform and
clay‐free sandstone unit suggest that in future studies, a wider range of petrophysical
relationships or a larger uncertainty bound should be assumed. The 𝑛 and 𝜌𝑠 estimates
do not appear to show significant correlation with other properties that were measured,
making it difficult to constrain petrophysical relationships with more core samples. In
fact, compared with previous studies at Hatfield and Eggborough, the use of more core
data reveals greater petrophysical model uncertainty. The individual Archie model fits
are good, but the concatenated data set shows a U‐shaped 𝜃(𝜌) behavior, which
suggests saturation is controlled by properties other than a saturation exponent or it
implies a heterogeneous petrophysical parameter field.
4.2. The uncertainty propagation approach
We have proposed and demonstrated an effective procedure to propagate
uncertainties in petrophysical relationships to uncertainties in the inferred moisture
contents and the amount of water within the plume. The procedure requires mean and
standard deviation of both the petrophysical parameters and the inverted resistivities.
The application of this method on field data using two types of petrophysical models
shows how uncertainty in petrophysical parameters and ERT data errors propagate
through the modeling and inversion process and lead to uncertainty in moisture
content estimates. Specifically, the inversion procedure smooths the resistivity profiles
(a proxy of moisture content) spatially, while the uncertain petrophysical relationships
add uncertainties to the quantitative conversion from resistivity to moisture content.
These uncertainties, if untracked, can lead to significant bias and over‐confidence in
the moisture content estimates.
Part of our analysis has utilized a commonly employed smoothness‐based
inversion for our geophysical data to evaluate the impact of uncertain petrophysical
Discussion and implications for future research
232
relationship. Other inversion algorithms may yield different uncertainty estimates. In
fact, a limitation of this work is that our computation of the uncertainty contribution
from inverted resistivity only considered the propagation of data errors through the
inversion code. We have assumed no uncertainty contribution from the choice of the
inverse model, its resolution, or its discretization, mainly because there is no standard
procedure to compute the uncertainty of an inverted resistivity field yet. Some
emerging techniques, such as trans‐dimensional ERT (Galetti and Curtis, 2018), are
attempts to address this issue. We also acknowledge Markov chain Monte Carlo
sampling (Brunetti and Linde, 2018) may be more accurate and robust than the
conventional MC sampling we use here.
Finally, we note that our approach follows the classical approach to error
analysis (Taylor, 1982). The extent to which some of the underlying assumptions are
valid, such as whether the uncertainties of petrophysical parameters and inverted
resistivities are independent, is open to future investigation. Nevertheless, we
highlight that the uncertainty propagation framework presented in this work is flexible
and straightforward. It is potentially applicable to any type of petrophysical models
and inversion methods, and it may be extended to consider the uncertainty of the
inversion itself. It is independent of the inversion methods and petrophysical models
used, and we expect it to be used widely in future studies.
4.3. Total moisture content estimation
The great variety of petrophysical models lead to a large range of total water
volume estimates (Figure 6a). This shows that using only a single petrophysical model
deterministically can give misleading results. It also shows that any applications
wishing to quantify the absolute amount of moisture present must not rely on
geophysics alone. The changes in moisture content estimated by Archie's law, however,
are generally consistent (Figure 6b). This can be explained by the work of Lehmann et
al. (2013), who show that the fractional changes in moisture content obtained from
electrical resistivity are a scaling of the saturation exponent only. This means the other
parameters in simple empirical models do not play a role in converting ratios of
inverted resistivities to ratios of 𝜃. Nevertheless, most applications are interested in at
Discussion and implications for future research
233
least the difference of moisture content between two times, not just their relative
change. We note the high uncertainty bounds associated with the change in 𝜃 obtained
from most of the Archie petrophysical models. This shows that this scaling of 𝑛 can
lead to highly uncertain estimates of the amount of the change. This effect should be
acknowledged and assumed when interpreting ERT‐derived moisture contents.
Moreover, other parameters in petrophysical models are still important in other
frequently used methods. For example, coupled modeling of hydrogeophysics
requires reliable petrophysical relationships. Examining the impact of the different
uncertain petrophysical parameters and models remains an important research
question.
Our uncertainty analysis shows that for most cases, the uncertainty in ERT‐
derived saturation is dominated by uncertain petrophysical parameters, not uncertain
inverted resistivities due to data errors (Figure 6c). This presents a challenge because
unlike inverted resistivities, petrophysical uncertainties cannot be straightforwardly
reduced by good quality data or better inverse modeling approaches. Future studies
should focus their efforts on better characterizing petrophysical uncertainties and
incorporating them in moisture content estimation procedures. Figure 6d also shows
that significant additional uncertainty can be caused by uncertain porosity values.
Since porosity ultimately controls the volume of pore space for water to fill, better
characterization of it can reduce the uncertainty of the moisture content estimates from
ERT.
Our work has focused on a water injection experiment where there is no
variation in fluid conductivity over time. Changes in fluid conductivity (e.g., in a saline
tracer injection or leak of saline solute) will further complicate the estimation of
moisture content changes since bulk resistivity is affected by both fluid conductivity
and moisture content. When inverting time‐lapse ERT data, the change relative to
baseline is often set to be minimized. This setting works well in our water injection
experiment but may give an insufficient change in resistivity to account for both
changes in saturation and fluid conductivity.
Discussion and implications for future research
234
Since laboratory petrophysical measurements are labor‐intensive and time‐
consuming, many authors have used TDR data (in shallow vadose zone investigations)
to fit field‐based petrophysical relationships (e.g. Fan et al., 2015). The typical setup,
for shallow investigations, consists of a trench with ERT, TDR, and temperature
sensors installed. This in situ setup can be viewed as advantageous over lab
measurements since it correctly represents pore water conductivity (given dynamic
exchange of ions between particles and pore water) and avoids forced conditions in
the lab. Despite its advantages, the range of 𝜌 it considers is limited because only the
range of the ERT‐measured apparent ρ are evaluated. Given the large variability of
petrophysical relationships observed in this study, perhaps the TDR data are best used
to independently verify or constrain the inverted moisture contents (e.g. Beff et al.,
2013). It is important to check independently whether the uncertainty bounds of ERT‐
predicted moisture content consistently capture the TDR data. While TDR or neutron
probe can only be applied in shallow soil, radar can be used in deeper investigations.
The joint use of ERT and radar measurements (e.g. Binley et al., 2002a; Linde et al.,
2006) yields independent estimates of moisture contents and allows cross‐validation.
We have examined the changes in total moisture content in the extracted
volume and at selected locations obtained from ERT and their agreement with the
simulation. Future uncertainty studies should consider the agreement by comparing
ERT estimates and other (e.g., TDR and neutron probe) data in the field. Further work
should also examine the extent to which the uncertainty in ERT‐derived moisture
content affects the decision making in specific applications, such as landslide
monitoring or precision agriculture.
4.4. Strategy when petrophysical data is unavailable
With the increasing popularity of ERT or EMI studies for hydrological
investigations, there will be an increasing number of studies that do not collect samples
for petrophysical calibration, which is often more time‐consuming than the
geophysical survey itself. Conversely, a few depth profiles of grain size distributions
are relatively easy to obtain (e.g., using a hand auger) in near‐surface applications. Soil
texture is commonly used to approximate unsaturated zone parameters through
Conclusion
235
pedotransfer functions (e.g., ROSETTA: Schaap et al., 2001; Zhang et al., 2016), and it
will be useful if these functions can approximate the petrophysical parameters or
models too. Future efforts should be devoted to building a global database on 𝜃(𝜌) and
grain size distribution data, in order to formulate pedotransfer functions across sites.
Data‐driven methods such as multiple adaptive regression splines (Brillante et al., 2014)
are particularly suitable for this task because they are capable of handling fairly large
datasets (e.g., 105 observations and 100 variables). We attempted to apply some of
these methods to fit the Eggborough data (not reported here), but we have too few
samples to apply them reliably. Nevertheless, they are potentially powerful methods
to apply in the future once there is a database for near‐surface petrophysical
measurements.
4.5. Relevance to EMI and other geophysical methods
We have focused mainly on the effect on ERT inversion results, but similar
conclusions can be extended for EMI results or methods that use a combination of EMI
and ERT results (von Hebel et al., 2014), as well as other applications in
hydrogeophysics where petrophysical transforms are involved. Moreover, we
recognize that there is a wealth of literature studying the spatial and temporal patterns
of electrical conductivity and soil moisture in the Earth's near‐surface. Similarly, there
have been many recent studies on data assimilation of moisture content data across
multiple spatial scales (e.g. Zhu et al., 2017). Hydrogeophysicists, while frequently
working at the plot‐scale and site‐scale, should be involved in these developments.
Closer collaboration between soil scientists, geostatisticians, geophysicists, and
hydrologists are needed to tackle this grand challenge.
5) Conclusion
Our study showed the extent of petrophysical variability present at a field site
and demonstrated an approach to computing uncertainty bounds of moisture content
estimates due to uncertain petrophysical models. First, we showed that highly variable
petrophysical relationships can be observed in field samples of a relatively uniform
and clay‐free sandstone unit. We then fitted and applied various petrophysical models
Acknowledgements
236
to convert ERT images to moisture content images. The different petrophysical models
led to a wide range of total moisture content estimates of a plume, but their changes
over time generally agreed. Using rules of error propagation, we were able to quantify
the uncertainty bounds of the moisture content estimates and gained further insight
by showing the individual contribution of the petrophysical parameters and inverted
resistivities terms to the total uncertainty. We showed that, assuming the inverse
model only smooths the resistivity field, the uncertainty is dominated by the
petrophysical parameters. The total uncertainty was found to be 7.52–23.18% of the
mean total water volume estimate. When translated to the change in time, the
uncertainty can be as high as several multiples of the mean estimate—both
uncertainties are higher than previously appreciated.
Our results have highlighted the potential danger of converting ERT images to
moisture content from similar environments using a single petrophysical model
deterministically. In particular, they should not be used to quantify the amount of
moisture present independently of other data. Although the different Archie
petrophysical models give consistent estimates of the change in total water volume,
their relatively large uncertainty bounds highlight that even though electrical
geophysics reliably determines the direction of the change in 𝜃, its quantification of the
amount of such change is highly uncertain. It is prudent to assume large uncertainties
for 𝜃 and 𝛥𝜃 estimates where they have not been quantified. Data‐driven methods (e.g.,
multiple adaptive regression splines) have the potential to be applied to build
petrophysical models where such data are unavailable.
6) Acknowledgements
This paper is published with the permission of the Executive Director of the British
Geological Survey (NERC) and UK Nuclear Decommissioning Authority (NDA). The
first author is supported by a Lancaster University Faculty of Science and Technology
PhD studentship and an NDA PhD Bursary. Peter Winship (deceased) and Melanie
Fukes conducted the field experiments and the laboratory analysis, respectively. We
thank Jared West (Leeds University) for supplying the Hatfield core analysis. We thank
References
237
Tim Johnson (Pacific Northwest National Laboratory) for assistance in using
PFLOTRAN‐E4D, access to the PNNL computing facilities, and hosting the first
author's visits. We thank Keith Beven for commenting on an initial version of the
manuscript. We are grateful to associate editor Sander Huisman, Sarah Garré, and an
anonymous reviewer for their constructive comments. The data used in this paper are
deposited at the Lancaster University's research data repository and is available for
download (https://dx.doi.org/10.17635/lancaster/researchdata/293).
References
Ahmed, M.A., Passioura, J., Carminati, A., 2018. Hydraulic processes in roots and the
rhizosphere pertinent to increasing yield of water-limited grain crops: a critical
review. J. Exp. Bot. 69, 3255–3265. https://doi.org/10.1093/jxb/ery183
Altdorff, D., von Hebel, C., Borchard, N., van der Kruk, J., Bogena, H.R., Vereecken,
H., Huisman, J.A., 2017. Potential of catchment-wide soil water content
prediction using electromagnetic induction in a forest ecosystem. Environ. Earth
Sci. 76, 111. https://doi.org/10.1007/s12665-016-6361-3
Amidu, S.A., Dunbar, J.A., 2007. Geoelectric Studies of Seasonal Wetting and Drying
of a Texas Vertisol. Vadose Zo. J. 6, 511. https://doi.org/10.2136/vzj2007.0005
Aster, R., Borchers, B., Thurber, C.H., 2005. Parameter estimation and inverse problems.
Elsevier, Burlington, Massachusetts, USA.
Bass, B., Cardenas, M.B., Befus, K.M., 2017. Seasonal Shifts in Soil Moisture throughout
a Semiarid Hillslope Ecotone during Drought: A Geoelectrical View. Vadose Zo.
J. 0, 0. https://doi.org/10.2136/vzj2016.11.0108
Beff, L., Günther, T., Vandoorne, B., Couvreur, V., Javaux, M., 2013. Three-dimensional
monitoring of soil water content in a maize field using Electrical Resistivity
Tomography. Hydrol. Earth Syst. Sci. 17, 595–609. https://doi.org/10.5194/hess-
17-595-2013
Binley, A., 2003. Vadose Zone Tracer Testing in the UK Sherwood Sandstone :
Hydrogeophysical Data Report. Livermore, California.
Binley, A., Beven, K., 2003. Vadose zone flow model uncertianty as conditioned on
geophysical data. Groundwater 41, 119–127.
References
238
Binley, A., Cassiani, G., Middleton, R., Winship, P., 2002a. Vadose zone flow model
parameterisation using cross-borehole radar and resistivity imaging. J. Hydrol.
267, 147–159. https://doi.org/10.1016/S0022-1694(02)00146-4
Binley, A., Cassiani, G., Winship, P., 2004. Characterization of heterogeneity in
unsaturated sandstone using borehole logs and cross-borehole tomography, in:
Aquifer Characterization, in: Aquifer Characterization. SEPM (Society for
Sedimentary Geology), pp. 129–138. https://doi.org/10.2110/pec.04.80.0129
Binley, A., Hubbard, S.S., Huisman, J.A., Revil, A., Robinson, D.A., Singha, K., Slater,
L.D., 2015. The emergence of hydrogeophsyics for improved understanding of
subsurface processes over multiple scales. Water Resour. Res. 51, 3837–3866.
https://doi.org/10.1002/2015WR017016
Binley, A., Slater, L.D., Fukes, M., Cassiani, G., 2005. Relationship between spectral
induced polarization and hydraulic properties of saturated and unsaturated
sandstone. Water Resour. Res. 41, W12417.
https://doi.org/10.1029/2005WR004202
Binley, A., Winship, P., Middleton, R., Pokar, M., West, J., 2001. High-resolution
characterization of vadose zone dynamics using cross-borehole radar. Water
Resour. Res. 37, 2639–2652. https://doi.org/10.1029/2000WR000089
Binley, A., Winship, P., West, L.J., Pokar, M., Middleton, R., 2002b. Seasonal variation
of moisture content in unsaturated sandstone inferred from borehole radar and
resistivity profiles. J. Hydrol. 267, 160–172. https://doi.org/10.1016/S0022-
1694(02)00147-6
Bouchedda, A., Chouteau, M., Binley, A., Giroux, B., 2012. 2-D joint structural
inversion of cross-hole electrical resistance and ground penetrating radar data. J.
Appl. Geophys. 78, 52–67. https://doi.org/10.1016/j.jappgeo.2011.10.009
Brillante, L., Bois, B., Mathieu, O., Bichet, V., Michot, D., Lévêque, J., 2014. Monitoring
soil volume wetness in heterogeneous soils by electrical resistivity. A field-based
pedotransfer function. J. Hydrol. 516, 56–66.
https://doi.org/10.1016/j.jhydrol.2014.01.052
Brillante, L., Mathieu, O., Bois, B., Van Leeuwen, C., Lévêque, J., 2015. The use of soil
electrical resistivity to monitor plant and soil water relationships in vineyards.
Soil 1, 273–286. https://doi.org/10.5194/soil-1-273-2015
Brunet, P., Clément, R., Bouvier, C., 2010. Monitoring soil water content and deficit
using Electrical Resistivity Tomography (ERT) - A case study in the Cevennes
area, France. J. Hydrol. 380, 146–153. https://doi.org/10.1016/j.jhydrol.2009.10.032
References
239
Brunetti, C., Linde, N., 2018. Impact of petrophysical uncertainty on Bayesian
hydrogeophysical inversion and model selection. Adv. Water Resour. 111, 346–
359. https://doi.org/10.1016/j.advwatres.2017.11.028
Cassiani, G., Binley, A., 2005. Modeling unsaturated flow in a layered formation under
quasi-steady state conditions using geophysical data constraints. Adv. Water
Resour. 28, 467–477. https://doi.org/10.1016/j.advwatres.2004.12.007
Cassiani, G., Godio, A., Stocco, S., Villa, A., Deiana, R., Frattini, P., Rossi, M., 2009.
Monitoring the hydrologic behaviour of a mountain slope via time-lapse
electrical resistivity tomography. Near Surf. Geophys. 7, 475–486.
https://doi.org/10.3997/1873-0604.2009013
Chambers, J.E., Wilkinson, P.B., Uhlemann, S., Sorensen, J.P.R., Roberts, C., Newell,
A.J., Ward, W.O.C., Binley, A., Williams, P.J., Gooddy, D.C., Old, G., Bai, L., 2014.
Derivation of lowland riparian wetland deposit architecture using geophysical
image analysis and interface detection. Water Resour. Res. 50, 5886–5905.
https://doi.org/10.1002/2014WR015643
Chen, H.C., Fang, J.H., 1986. Sensitivity analysis of the parameters in Archie’s water
saturation equation. Log Anal. 27, 39–44. https://doi.org/SPWLA-1986-
vXXVIIn5a3
Doetsch, J., Linde, N., Vogt, T., Binley, A., Green, A.G., 2012. Imaging and quantifying
salt-tracer transport in a riparian groundwater system by means of 3D ERT
monitoring. Geophysics 77, B207. https://doi.org/10.1190/geo2012-0046.1
Fan, J., Scheuermann, A., Guyot, A., Baumgartl, T., Lockington, D.A., 2015.
Quantifying spatiotemporal dynamics of root-zone soil water in a mixed forest
on subtropical coastal sand dune using surface ERT and spatial TDR. J. Hydrol.
523, 475–488. https://doi.org/10.1016/j.jhydrol.2015.01.064
Friedman, S.P., 2005. Soil properties influencing apparent electrical conductivity: a
review. Comput. Electron. Agric. 46, 45–70.
https://doi.org/10.1016/j.compag.2004.11.001
Galetti, E., Curtis, A., 2018. Transdimensional Electrical Resistivity Tomography. J.
Geophys. Res. Solid Earth. https://doi.org/10.1029/2017JB015418
Garré, S., Coteur, I., Wongleecharoen, C., Kongkaew, T., Diels, J., Vanderborght, J.,
2013. Noninvasive Monitoring of Soil Water Dynamics in Mixed Cropping
Systems: A Case Study in Ratchaburi Province, Thailand. Vadose Zo. J. 12, 0.
https://doi.org/10.2136/vzj2012.0129
References
240
Garré, S., Javaux, M., Vanderborght, J., Pagès, L., Vereecken, H., 2011. Three-
Dimensional Electrical Resistivity Tomography to Monitor Root Zone Water
Dynamics. Vadose Zo. J. 10, 412. https://doi.org/10.2136/vzj2010.0079
Glover, P.W.J., 2017. A new theoretical interpretation of Archie’s saturation exponent.
Solid Earth 8, 805–816. https://doi.org/10.5194/se-8-805-2017
Glover, P.W.J., 2015. Geophysical Properties of the Near Surface Earth: Electrical
Properties, in: Treatise on Geophysics. Elsevier, pp. 89–137.
https://doi.org/10.1016/B978-0-444-53802-4.00189-5
Green, C.T., Liao, L., Nolan, B.T., Juckem, P.F., Shope, C.L., Tesoriero, A.J., Jurgens,
B.C., 2018. Regional Variability of Nitrate Fluxes in the Unsaturated Zone and
Groundwater, Wisconsin, USA. Water Resour. Res. 54, 301–322.
https://doi.org/10.1002/2017WR022012
Haarder, E.B., Binley, A., Looms, M.C., Doetsch, J., Nielsen, L., Jensen, K.H., 2012.
Comparing Plume Characteristics Inferred from Cross-Borehole Geophysical
Data. Vadose Zo. J. 11, 1–40. https://doi.org/10.2136/vzj2012.0031
Hammond, G.E., Lichtner, P.C., Mills, R.T., 2014. Evaluating the performance of
parallel subsurface simulators: An illustrative example with PFLOTRAN. Water
Resour. Res. 50, 208–228. https://doi.org/10.1002/2012WR013483
Hen-Jones, R.M., Hughes, P.N., Stirling, R.A., Glendinning, S., Chambers, J.E., Gunn,
D.A., Cui, Y.J., 2017. Seasonal effects on geophysical–geotechnical relationships
and their implications for electrical resistivity tomography monitoring of slopes.
Acta Geotech. 12, 1159–1173. https://doi.org/10.1007/s11440-017-0523-7
Hermans, T., Nguyen, F., Caers, J., 2015. Uncertainty in training image-based inversion
of hydraulic head data constrained to ERT data:Workflow and case study. Water
Resour. Res. 51, 5332–5352. https://doi.org/10.1002/2014WR016259
Hinnell, A.C., Ferr, T.P.A., Vrugt, J.A., Huisman, J.A., Moysey, S., Rings, J., Kowalsky,
M.B., 2010. Improved extraction of hydrologic information from geophysical
data through coupled hydrogeophysical inversion. Water Resour. Res. 46, 1–14.
https://doi.org/10.1029/2008WR007060
Hübner, R., Heller, K., Günther, T., Kleber, A., 2015. Monitoring hillslope moisture
dynamics with surface ERT for enhancing spatial significance of hydrometric
point measurements. Hydrol. Earth Syst. Sci. 19, 225–240.
https://doi.org/10.5194/hess-19-225-2015
References
241
Jayawickreme, D.H., van Dam, R.L., Hyndman, D.W., 2008. Subsurface imaging of
vegetation, climate, and root-zone moisture interactions. Geophys. Res. Lett. 35,
1–5. https://doi.org/10.1029/2008GL034690
Johnson, T.C., Hammond, G.E., Chen, X., 2017. PFLOTRAN-E4D: A parallel open
source PFLOTRAN module for simulating time-lapse electrical resistivity data.
Comput. Geosci. 99, 72–80. https://doi.org/10.1016/j.cageo.2016.09.006
Johnson, T.C., Versteeg, R.J., Ward, A., Day-Lewis, F.D., Revil, A., 2010. Improved
hydrogeophysical characterization and monitoring through parallel modeling
and inversion of time-domain resistivity and induced-polarization data.
Geophysics 75, WA27. https://doi.org/10.1190/1.3475513
Jonard, F., Bogena, H., Caterina, D., Garré, S., Klotzsche, A., Monerris, A., Schwank,
M., von Hebel, C., 2018. Ground-Based Soil Moisture Determination, in: Li, X.,
Vereecken, H. (Eds.), Observation and Measurement of Ecohydrological
Processes. Springer, pp. 1–42. https://doi.org/10.1007/978-3-662-47871-4_2-1
Koestel, J., Kemna, A., Javaux, M., Binley, A., Vereecken, H., 2008. Quantitative
imaging of solute transport in an unsaturated and undisturbed soil monolith
with 3-D ERT and TDR. Water Resour. Res. 44, 1–17.
https://doi.org/10.1029/2007WR006755
LaBrecque, D.J., Heath, G., Sharpe, R., Versteeg, R., 2004. Autonomous Monitoring of
Fluid Movement Using 3-D Electrical Resistivity Tomography. J. Environ. Eng.
Geophys. 9, 167–176. https://doi.org/10.4133/JEEG9.3.167
Labrecque, D.J., Yang, X., 2001. Difference inversion of ERT data: a fast inversion
method for 3-D in situ monitoring. J. Environ. Eng. Geophys. 6, 83–89.
https://doi.org/dx.doi.org/10.4133/JEEG6.2.83
Laloy, E., Javaux, M., Vanclooster, M., Roisin, C., Bielders, C.L., 2011. Electrical
Resistivity in a Loamy Soil: Identification of the Appropriate Pedo-Electrical
Model. Vadose Zo. J. 10, 1023. https://doi.org/10.2136/vzj2010.0095
Lehmann, P., Gambazzi, F., Suski, B., Baron, L., Askarinejad, A., Springman, S.M.,
Holliger, K., Or, D., 2013. Evolution of soil wetting patterns preceding a
hydrologically induced landslide inferred from electrical resistivity survey and
point measurements of volumetric water content and pore water pressure. Water
Resour. Res. 49, 7992–8004. https://doi.org/10.1002/2013WR014560
Lesmes, D.P., Friedman, S.P., 2005. Relationships between the Electrical and
Hydrogeological Properties of Rocks and Soils, in: Rubin, Y., Hubbard, S.S. (Eds.),
References
242
Hydrogeophysics. Springer Netherlands, Dordrecht, pp. 87–128.
https://doi.org/10.1007/1-4020-3102-5_4
Linde, N., Binley, A., Tryggvason, A., Pedersen, L.B., Revil, A., 2006. Improved
hydrogeophysical characterization using joint inversion of cross-hole electrical
resistance and ground-penetrating radar traveltime data. Water Resour. Res. 42,
WR005131. https://doi.org/10.1029/2006WR005131
Looms, M.C., Binley, A., Jensen, K.H., Nielsen, L., Hansen, T.M., 2008. Identifying
Unsaturated Hydraulic Parameters Using an Integrated Data Fusion Approach
on Cross-Borehole Geophysical Data. Vadose Zo. J. 7, 238.
https://doi.org/10.2136/vzj2007.0087
Martini, E., Werban, U., Zacharias, S., Pohle, M., Dietrich, P., Wollschläger, U., 2017.
Repeated electromagnetic induction measurements for mapping soil moisture at
the field scale: validation with data from a wireless soil moisture monitoring
network. Hydrol. Earth Syst. Sci. 21, 495–513. https://doi.org/10.5194/hess-21-
495-2017
Michot, D., Benderitter, Y., Dorigny, A., Nicoullaud, B., King, D., Tabbagh, A., 2003.
Spatial and temporal monitoring of soil water content with an irrigated corn crop
cover using surface electrical resistivity tomography. Water Resour. Res. 39, 1–
20. https://doi.org/10.1029/2002WR001581
Miller, C.R., Routh, P.S., Brosten, T.R., McNamara, J.P., 2008. Application of time-lapse
ERT imaging to watershed characterization. Geophysics 73, G7.
https://doi.org/10.1190/1.2907156
Oldenborger, G.A., LeBlanc, A.M., 2015. Geophysical characterization of permafrost
terrain at Iqaluit International Airport, Nunavut. J. Appl. Geophys. 123, 36–49.
https://doi.org/10.1016/j.jappgeo.2015.09.016
Ray, R.L., Jacobs, J.M., 2007. Relationships among remotely sensed soil moisture,
precipitation and landslide events. Nat. Hazards 43, 211–222.
https://doi.org/10.1007/s11069-006-9095-9
Robinet, J., von Hebel, C., Govers, G., van der Kruk, J., Minella, J.P.G., Schlesner, A.,
Ameijeiras-Mariño, Y., Vanderborght, J., 2018. Spatial variability of soil water
content and soil electrical conductivity across scales derived from
Electromagnetic Induction and Time Domain Reflectometry. Geoderma 314,
160–174. https://doi.org/10.1016/j.geoderma.2017.10.045
Robinson, D.A., Lebron, I., Kocar, B., Phan, K., Sampson, M., Crook, N., Fendorf, S.,
2009. Time-lapse geophysical imaging of soil moisture dynamics in tropical
References
243
deltaic soils: An aid to interpreting hydrological and geochemical processes.
Water Resour. Res. 45, n/a-n/a. https://doi.org/10.1029/2008WR006984
Scaini, A., Audebert, M., Hissler, C., Fenicia, F., Gourdol, L., Pfister, L., Beven, K.J.,
2017. Velocity and celerity dynamics at plot scale inferred from artificial tracing
experiments and time-lapse ERT. J. Hydrol. 546, 28–43.
https://doi.org/10.1016/J.JHYDROL.2016.12.035
Schaap, M.G., Leij, F.J., van Genuchten, M.T., 2001. rosetta: a computer program for
estimating soil hydraulic parameters with hierarchical pedotransfer functions. J.
Hydrol. 251, 163–176. https://doi.org/10.1016/S0022-1694(01)00466-8
Scholer, M., Irving, J., Binley, A., Holliger, K., 2011. Estimating vadose zone hydraulic
properties using ground penetrating radar: The impact of prior information.
Water Resour. Res. 47, 1–14. https://doi.org/10.1029/2011WR010409
Shanahan, P.W., Binley, A., Whalley, W.R., Watts, C.W., 2015. The Use of
Electromagnetic Induction to Monitor Changes in Soil Moisture Profiles beneath
Different Wheat Genotypes. Soil Sci. Soc. Am. J.
https://doi.org/10.2136/sssaj2014.09.0360
Si, H., 2015. TetGen, a Delaunay-Based Quality Tetrahedral Mesh Generator. ACM
Trans. Math. Softw. 41, 1–36. https://doi.org/10.1145/2629697
Singha, K., Day-Lewis, F.D., Johnson, T., Slater, L.D., 2015. Advances in interpretation
of subsurface processes with time-lapse electrical imaging. Hydrol. Process. 29,
1549–1576. https://doi.org/10.1002/hyp.10280
Taylor, J.R., 1982. An Introduction to Error Analysis: The Study of Uncertainties in
Physical Measurements. University Science Books.
Tso, C.-H.M., Kuras, O., Wilkinson, P.B., Uhlemann, S., Chambers, J.E., Meldrum, P.I.,
Graham, J., Sherlock, E.F., Binley, A., 2017. Improved characterisation and
modelling of measurement errors in electrical resistivity tomography (ERT)
surveys. J. Appl. Geophys. 146, 103–119.
https://doi.org/10.1016/J.JAPPGEO.2017.09.009
Turkeltaub, T., Jia, X., Zhu, Y., Shao, M.-A., Binley, A., 2018. Recharge and Nitrate
Transport Through the Deep Vadose Zone of the Loess Plateau: A Regional-Scale
Model Investigation. Water Resour. Res. 54, 4332–4346.
https://doi.org/10.1029/2017WR022190
Uhlemann, S., Chambers, J., Wilkinson, P., Maurer, H., Merritt, A., Meldrum, P., Kuras,
O., Gunn, D., Smith, A., Dijkstra, T., 2017. Four-dimensional imaging of moisture
References
244
dynamics during landslide reactivation. J. Geophys. Res. Earth Surf. 122, 398–
418. https://doi.org/10.1002/2016JF003983
Uhlemann, S., Sorensen, J.P.R., House, A.R., Wilkinson, P.B., Roberts, C., Gooddy, D.C.,
2016. Integrated time-lapse geoelectrical imaging of wetland hydrological
processes. Water Resour. Res. 52, 1607–1625.
https://doi.org/10.1002/2015WR017932
von Hebel, C., Rudolph, S., Mester, A., Huisman, J.A., Kumbhar, P., Vereecken, H., van
der Kruk, J., 2014. Three-dimensional imaging of subsurface structural patterns
using quantitative large-scale multiconfiguration electromagnetic induction data.
Water Resour. Res. 50, 2732–2748. https://doi.org/10.1002/2013WR014864
Wehrer, M., Slater, L.D., 2015. Characterization of water content dynamics and tracer
breakthrough by 3-D electrical resistivity tomography (ERT) under transient
unsaturated conditions. Water Resour. Res. 51, 97–124.
https://doi.org/10.1002/2014WR016131
Weller, A., Slater, L., Nordsiek, S., 2013. On the relationship between induced
polarization and surface conductivity: Implications for petrophysical
interpretation of electrical measurements. Geophysics 78, D315–D325.
https://doi.org/10.1190/geo2013-0076.1
Whalley, W.R., Binley, A., Watts, C.W., Shanahan, P., Dodd, I.C., Ober, E.S., Ashton,
R.W., Webster, C.P., White, R.P., Hawkesford, M.J., 2017. Methods to estimate
changes in soil water for phenotyping root activity in the field. Plant Soil 415,
407–422. https://doi.org/10.1007/s11104-016-3161-1
Wiese, B., Wagner, F., Norden, B., Maurer, H., Schmidt-Hattenberger, C., 2018. Fully
coupled inversion on a multi-physical reservoir model - Part I: Theory and
concept. Int. J. Greenh. Gas Control.
Winship, P., Binley, A., Gomez, D., 2006. Flow and transport in the unsaturated
Sherwood Sandstone: characterization using cross-borehole geophysical
methods. Geol. Soc. London, Spec. Publ. 263, 219–231.
https://doi.org/10.1144/GSL.SP.2006.263.01.12
Yamakawa, Y., Kosugi, K., Katsura, S., Masaoka, N., Mizuyama, T., 2012. Spatial and
Temporal Monitoring of Water Content in Weathered Granitic Bedrock Using
Electrical Resistivity Imaging. Vadose Zo. J. 11, 0.
https://doi.org/10.2136/vzj2011.0029
Appendix A: petrophysical uncertainty propagation
245
Yonggen, Z., Schaap, M.G., Guadagnini, A., Neuman, S.P., 2016. Inverse modeling of
unsaturated flow using clusters of soil texture and pedotransfer functions. Water
Resour. Res. 52, 7631–7644. https://doi.org/10.1002/2016WR019016
Zhu, P., Shi, L., Zhu, Y., Zhang, Q., Huang, K., Williams, M., 2017. Data assimilation of
soil water flow via ensemble Kalman filter: Infusing soil moisture data at
different scales. J. Hydrol. 555, 912–925.
https://doi.org/10.1016/j.jhydrol.2017.10.078
Appendix A: petrophysical uncertainty propagation
Following the analytical sensitivity analysis of Chen and Fang (1986) and Taylor (1982),
we can obtain the uncertainty contributions of the various terms in Archie's law
(equation 2). Assuming they have uncorrelated errors, by laws of error propagation,
the variance of saturation is given by
𝜎𝑆𝑤2 = (
𝜕𝑆𝑤𝜕𝜌𝑠
)2
𝜎𝜌𝑠2 + (
𝜕𝑆𝑤𝜕𝜌)2
𝜎𝜌2 + (
𝜕𝑆𝑤𝜕𝑛)2
𝜎𝑛2
where
𝜕𝑆𝑤𝜕𝜌𝑠
=1
𝑛𝜌𝑠𝑆𝑤
𝜕𝑆𝑤𝜕𝜌
=1
𝑛𝜌𝑆𝑤
𝜕𝑆𝑤𝜕𝑛
=ln (𝜌 𝜌𝑠⁄ )
𝑛2𝑆𝑤
So
(𝜕𝑆𝑤𝜕𝜌𝑠
)2
𝜎𝜌𝑠2 = (
𝑆𝑤𝑛)2
(𝜎𝜌𝑠𝜌𝑠)2
(𝜕𝑆𝑤𝜕𝜌)2
𝜎𝜌2 = (
𝑆𝑤𝑛)2
(𝜎𝜌
𝜌)
2
(𝜕𝑆𝑤𝜕𝑛)2
𝜎𝑛2 = (
𝑆𝑤𝑛)2
(ln (𝜌/𝜌𝑠)
𝑛𝜎𝑛)
2
Supplementary information
Text S1.
Supplementary information
246
Monte Carlo ERT inversion runs
Given the data errors of the ERT measurements, the Monte Carlo uncertainty
propagation procedure of Aster et al. (2005) (as used in Tso et al. (Tso et al., 2017)) can
be used and is outlined below:
8. Propagate the inverse solution into an assumed noise-free baseline jx1 data
vector 𝐝 (where j is the size of number of measurements) using the forward
model G:
𝐆 = 𝐝 (11)
9. Generate q realizations (i = 1, …, q) of noisy data about using the error
model
𝐝i = 𝐝b + 𝛆.∗ 𝐙 (12)
where 𝜺 is the j x1 vector of error levels predicted by the error model and Z is
the standard normal distribution variable and .* is element-wise multiplication.
10. Invert the q realizations (i = 1, …, q) of noisy data using the inverse model
𝐆𝐦i = 𝐝b + 𝛆i (13)
11. Let A be a q x m matrix where the i-th row contains the departure of the i-th
model from the baseline inverse solution
𝐀i = 𝐦iT − T (14)
12. An empirical estimate of the model covariance matrix is given by
cov() =𝐀T𝐀
𝑞 (15)
13. 95% confidence interval about is given by
± 1.96 ∙ diag(cov())1/2 (16)
14. Similarly, the coefficient of variation of the estimate is given by
diag(cov())1/2./𝑇 (17)
where ./ is element-wise division.
Supplementary information
247
Figure S1. Inverted ERT images from the Hatfield controlled infiltration experiment. The electrode locations are shown in purple squares. The iso-surfaces show area that has 5.5% reduction of resistivity relative to baseline (Day 7).
Table S1. Relationship between Archies parameters with clay content, cation exchange capacity (CEC), and surface area to pore volume ratio (Spor). Note that Spor data is not available for some cores.
R2 Clay content
(%)
CEC
(meq/100g)
Spor
(μm-1)
𝜌𝑠 (m) 0.2655 0.2913 0.6301
𝑛 (−) 0.0017 0.0871 0.2018
Discussion summary
248
7. Discussion summary
A primary goal of this thesis is to identify the sources of uncertainty in ERT
data collection, modelling, inversion, and interpretation. In the introduction section, I
outlined a 5-step workflow (Figure 1.1) for ERT and argued that information and
uncertainty propagates through it. I then identified three specific research areas to
address in chapters 4-6 of this thesis. I have reviewed in chapter 2 the geophysical
methods that have been used extensively for nuclear site characterisation (particularly
in the USA). Uncertainty and risk assessment is a recurring topic in these work because
of the management decisions that are made at these sites. A comprehensive review of
the sources of uncertainties in the ERT workflow was given in chapter 3, which
provides an overview of existing work on handling uncertainty at specific stages of the
workflow. Most of the work reviewed focused on inversion, with a few exceptions that
considers the remaining aspects. Summaries of the findings of this thesis can be found
below and in Figure 1.
Chapter 4 reviewed the various aspects of measurement errors in ERT. I
compared the three common type of errors obtained during data acquisition—stacking
errors, repeatability errors, and reciprocal errors and assessed the statistical
distribution and correlation. Using data from two long-term monitoring dataset
provided by the British Geological Survey, it was found that stacking errors are
consistently an order of magnitude lower than others while repeatability and
reciprocal errors tend to agree and show little auto-correlation. Since errors obtained
in the field is only an instance in the assumed Gaussian distribution for noise, they
need to be modelled before being fed to inversion. Linear models are commonly used
to account for proportionality effects. It was found that each electrode used for making
a ERT measurement has a secondary effect on errors. Such effect is incorporated using
a new linear mixed effect model and it leads to superior inversion results in cases
where certain electrodes are of significantly lower quality than others.
In some applications, the goal of using ERT is to predict some quantities of
interests while the ERT images themselves are not directly relevant. Chapter 5 assessed
the benefit of leak parameter estimation using ERT data directly (i.e. without inversion).
Discussion summary
249
It provided a first attempt to combine steps 3-5 of the ERT workflow (Figure 1, see also
chapter 1) to estimate leak parameters directly from raw ERT data. This can be
achieved by combining coupled hydrogeophysical modelling and data assimilation,
where the ERT data misfits update the distribution of flow model parameters. The
reduction of model uncertainty upon conditioning on ERT data can be visualised in
reduced spread in model parameters as well as mass discharge curves over a user-
defined plane. The proposed method is promising as it offers a way to estimate leak
parameters from ERT data, but care should be taken in model proposals and
interpreting the posterior. For leak detection and similar problems, the proposed
method can provide additional insights that complements ERT inversion. It is
particularly releveant to nuclear sites because most of them have undergone series of
site investigation over the past decades and has a wealth of prior site characterisation
information. It should be noted such framework is not limited to geophysical data but
can be used in any site characterisation work. The key idea is to use data to constrain
one’s prior understand of the site conceptualization.
Geophysics are also frequently used in the unsaturated zone to infer changes
in spatial distribution of moisture content with time. This is important because higher
moisture content can mean more water available for crop growth or faster solute
transport from near-surface sources to aquifers. However, the spatially varying
resolution of geophysical inversion as well as the uncertainties in the petrophysical
relationships relating moisture content and geophysical properties can complicate the
inference of moisture content from geophysical data. Chapter 6 addressed a classic
problem in the interpretation of ERT image—the quantification of an imaged moisture
plume. The chapter examined the extent to which uncertain petrophysical
relationships affect the estimation of soil water content (and its temporal changes).
These effects were quantified and compared for the first time using an uncertainty
propagation framework for resistivity-derived moisture content estimates. It was
shown that natural variability of petrophysical models can be high and as a
consequence the moisture content estimates can vary significantly. It was also shown
that it is possible that petrophysical parameters to be a more significant source of
uncertainty than ERT measurement errors and inversion. The original goal in chapter
Discussion summary
250
6 was to quantify the uncertainty in the volume of a solute plume. However, there is
no straightforward way to disentangle the contribution of saturation and solute
concentration to resistivity. An analysis like the one in chapter 6 to solutes is highly
relevant, but some major adaptations need to be made.
Figure 1 A diagram summarizing the major findings in this thesis and their relation to the ERT
workflow.
Long-term ERT monitoring has provided a new dimension to the wealth of
geophysical data available. For example, in the Sellafield data, daily data for about 1.5
years is available. An initial ambition that was not materialized is to make use of the
full time series of the ERT data such as to make "live" leak/no-leak calls based on raw
ERT data and to automatically flag “interesting” time slices for inversion. In this thesis,
however, full time series are only used in the measurement error characterisation
chapter. Others have only used data from a few selected time slices available in order
to save computation time (note that the observartion ensemble is ((𝑛𝑑 × 𝑛𝑡) × (𝑛𝑒),
where 𝑛𝑑, 𝑛𝑡 and 𝑛𝑒 are number of measurements, number of time slices, and size of
the ensemble respectively. Recently, I have come across statistical methods for
changepoint detection (i.e. mean and/or variance shifts in time series) (Killick et al.,
Discussion summary
251
2012) to automatically identify features of interest from ERT data with both online and
offline modes. It has been used in an increasing number of environmental applications
and they have potential to be used for geophysical monitoring data.
In this thesis, a deliberate decision was made not to focus on ERT inversion so
that other sources of uncertainties can be considered. However, inversion remains an
integral part of the ERT workflow and UQ in ERT and it will remain an important area
of research. To obtain inversion uncertainty, we have used traditional Monte Carlo
sampling for simplicity. There exist other alternatives such as Bootstrap inversion
(Schnaidt and Heinson, 2015) that can improve the efficiency, accuracy, and robustness
of such sampling. We also recognize that the uncertainty estimates from the above
procedures do not represent the uncertainty of the parameter space, but rather the
effect of parameter uncertainty due to data errors (i.e. how data errors are propagated
through inversion). Some of the inversion approaches listed in the following are
perhaps more suitable for this task.
Because Markov chain Monte Carlo (McMC) methods conditionally accept and
reject model proposals, it is argued that they provide more robust and accurate
uncertainty estimates. In particular, it provides the full posterior parameter probability
distribution, which can be multi-modal in highly uncertain regions such as interfaces.
At the start of the PhD, I decided not to use McMC because of its very high
computation cost. This high cost is in part due to computation needed to generate
model proposals, and in part due to parameterization (i.e. very fine parameterization
is used), among other factors. Significant progress has been made on McMC in both
areas. For example, graph cuts (Zahner et al., 2016) allow rapid generation of many
model proposals with the same characteristic features using training images. Trans-
dimensional ERT (Galetti and Curtis, 2018) allows estimation of unstructured
parameter cell sizes and shapes alongside with their resistivity values. Recently, a two-
step methodology based on area-to-point kriging is proposed to generate fine-scale
multi-Gaussian realizations from smooth tomographic images (Nussbaumer et al.,
2019).
Discussion summary
252
Even though there are many inversion strategies available, our understanding
on the topology of the ERT inverse problem or the parameter space uncertainty
behaviour is limited. One piece of work I have started during my PhD is to investigate
whether we can use gradient-based approach to locate regions of global minima. The
concept here is to randomly sample the parameter space with particles and calculate
their local gradient. Then starting with one of the particles, look for the next closest
particle that is downhill of the current one. The process will continue until the search
path reaches a valley, where it will either bounce between two particle locations, or it
will form an orbit that joins several particle locations that outlines the boundary of the
valley. An example is shown in Figure 2. This approach, first proposed by Curtis and
Spencer (1999), is not itself an inversion method, but it can improve the modeller’s
understanding of the structure of the inverse problem they are facing, with a much
smaller computational cost than an uncertainty analysis that uses a fully Bayesian
McMC inversion.
Although McMC inversions are promising and advance rapidly, more efficient
methods (e.g. 3-D problems that can be calculated on a laptop) for ERT inversion with
uncertainty quantification and more flexible parameterization to account for spatial
scales are needed to aid enhancing the information content of geophysical data and its
uncertainty quantification. My leak detection work is one of the first that uses data
assimilation for raw ERT data. Such methods should also be considered, where
appropriate, for ERT inversion. There is potential for data assimilation methods to be
combined with level set or multi-point geostatistics (MPS) parameterization to
estimate a multi-scale heterogeneous resistivity field efficiently and provides
approximation of its uncertainty. Such methods allow target features that do not have
a constant length scale and normally not well recovered by smoothness-constrained or
traditional covariance-based (i.e. two-point) geostatistical inversion to be resolved
reasonably well.
The simulation of some hydrological problems can require a very long run time,
which makes it prohibitive to use them directly for uncertainty quantification.
Examples include regional-scale modelling of surface water-groundwater exchange or
Discussion summary
253
predicting the fate of radionuclides decades or centuries into the future. In these cases,
building surrogate models that approximate the behaviour of the full forward model
will be useful. It ensures that the the uncertainty quantification methods described in
this thesis is applicable to all hydrogeophysical problems.
Big data and machine learning methods have brought about radical changes in
many fields. Application of data-driven methods has emerged in recent
hydrogeophysical studies (Laloy et al., 2017; Nguyen et al., 2016). Bayesian evidential
learning (Hermans et al., 2018) has also been applied, where it provides an incremental
framework for characterisation to inform site management decisions. Specifically,
before data acquisition, Monte Carlo simulations and global sensitivity analysis are
used to assess the information content of the data to reduce the uncertainty of the
prediction. After data acquisition, prior falsification and machine learning based on
the same Monte Carlo simulations are used to directly assess uncertainty on key
prediction variables from observations. A big challenge of applying machine learning
methods is the availability of training data. For example, I have attempted to train a
data-driven model for petrophysical relationships but the data I had access to was
insufficient to build a robust model. The hydrogeophysics community can work
together to build databases of laboratory and field data for building machine learning
models. Recent work in surface water hydrology has also explored the use of deep
learning methods (Shen, 2018), which has begun to show potential in
hydrogeophysical studies (Laloy et al., 2017).
Forty years ago, Lytle and Dines (1978) developed the so-called “impedance
camera”, which can be considered as one of the earliest work on ERT. They noted
“Items worthy of future research include an assessment of the influence of noise in the data, a
study of the accuracy of the reconstruction and its spatial dependence, and evaluation of the
degree of dependence of various measurement configurations, an analytic study of the resolution
limit, and a determination of the extent to which the use of a priori knowledge affects the
interpretation. ” As addressed in this thesis, the challenges these pioneers highlighted
are more far-reaching and relevant than they were then and 40 years on we are getting
there.
Discussion summary
254
Figure 2 (Top) Parameter space of a 3-parameter layered ERT problem using 24 surface electrodes in
dipole-dipole configuration. The axis represents the uniform parameter value of each layer in log scale.
The true resistivity for all three layers are 100 𝛀 𝐦. The red cross indicate the true parameter value. The
data misfit surface (left) and the derived streamlines to the data misfit mimima (right) are plotted.
(bottom) Gradient fields using 500 samples. The red polygons are the loops and the black cross is the
true values. The gradient field somewhat point towards the true minima. 9 loops are identified,
spanning a large fraction of the parameter space.
Conclusions and recommendations
255
8. Conclusions and recommendations
8.1 Conclusions
Geophysical methods have been a suite of valuable tools to provide
information for site characterisation. This thesis has formalised the ERT workflow as a
pipeline for information/uncertainty propagation. It has focused on the sources of
uncertainties in the various aspects of the ERT workflow while making contribution in
better understanding several of its less explored aspects.
Through literature review and new analysis included here, we now better
understand the behaviour of the various sources of uncertainties in ERT. We have also
investigated how uncertainties at a given stage of the workflow propagates
downstream. Using rules of error propagation, the impact of uncertainties that
stemmed at any stage of the workflow can be approximated using Monte Carlo
analysis. This provides a powerful design and evaluation framework to quantify, rank,
and in some cases reduce the sources of uncertainties in an ERT study.
Statistical analysis of ERT measurement errors from long-term monitoring
datasets shows that the use of reciprocal and repeatability errors are preferred and they
have negligible correlation in time. Variations in electrode quality may have a minor
effect on measurement errors and a new error model is proposed to account for such
effects. Similarly, detailed examination of field samples allows us to better understand
the variability of petrophysical relationships in real-world settings. The impact of such
variability to moisture content estimates as it is propagated down the ERT workflow
is then assessed, which is found to be consistent but with larger error bars than
previously appreciated. A method that combined the inversion and prediction steps
together to bypass the typical workflow is also developed. It is suitable for situations
where there is abundant prior site information and a clear list of non-geophysical
parameters to be estimated. This approach is applied to leak detection problems and
estimated leak parameters from ERT data (with the help of prior hydrogeological
model) directly, which is not achievable from typical ERT inversion.
Conclusions and recommendations
256
The findings in this thesis are highly relevant to nuclear site characterisation. A
literature review in the thesis shows geophysics is widely applied in nuclear sites and
the industry has been used to dealing with uncertainties. The methods developed in
the thesis not only can improve the reliability of using ERT in general but also provides
a wealth of information for decision making at a nuclear site.
This thesis contributes to a growing number of studies which aims to extract
additional meaningful information from ERT datasets (e.g. see PhD work of Crestani,
2013; Robinson, 2015; Wagner, 2016; Ward, 2018). The results from the work presented
here demonstrates both the capability to extract more information from ERT, while the
linked issue to better track and quantify uncertainties in ERT studies is highlighted.
This thesis presents an uncertainty propagation framework that can serve as a basis
for further development of methods that seek to extract additional information from
ERT data for site characterisation.
8.2 Future work
Due to the limited scope of this thesis, many of the aspects on the information
content and uncertainty for geophysical data for site characaterization have not been
fully investigated. An area of research that has not been addressed in this work is to
track and quantify the amount of information and uncertainty that propagate through
the entire ERT workflow. This can be investigated further using both synthetic
examples and real-world problems. Once we have estimated the contribution of
uncertainty and information in each step and the way they propagate throughout the
pipeline, we can optimize each step for experimental design. This thesis focuses on
ERT only, but the approach developed in it are applicable to other geophysics or site
characterisation methods. Future work should also investigate the uncertainty
propagation where there is fusion of data types and multiple modalities, e.g. fusion of
satellite, air-borne, surface, and ground-based measurements. Similar uncertainty
quantification methods can be extended the related concept of value of information
(VOI), which computes the expected monetary return of a characterisation option.
Such calculations can rank characterisation methods in terms of their cost-effectiveness
and show the added value of including each additional method. An alternative to VOI
Bibliography
257
calculations is information content calculations (JafarGandomi and Binley, 2013;
Nearing and Gupta, 2015), where the marginal gain in information from various
characterisation options can be compared. Finally, a vision has been proposed recently
to perform integrated modelling and uncertainty analysis in virtual laboratories using
cloud computing platforms (Blair et al., 2019, 2018). One use of collaborative
workspaces is that it allows users to perform individual modelling tasks and specify a
workflow to compute uncertainty propagation across all components, which is well-
suited for the ERT workflow described in this thesis.
This thesis has considered the behaviour and detection of a single leak event
with a single source of assumed constant loading under “clean” antecedent conditions.
Of course, these assumptions do not always hold in real-world applications. There is
limited existing work that consider multiple source identification, most of them are
based on Bayesian formulation. It is my hope that my work presented here can be
generalized to handle more complex conditions that are present in some of the most
complex environmental sites in the world.
Part of this thesis uses data assimilation methods on ERT data to estimate leak
parameters and uniform hydro geophysical parameters. Some recent work has used
data assimilation from head, concentration, or ERT data (in some cases, their
combination) to estimate hydraulic parameter fields or plume geometry. Future work
should explore the development of a more generalized framework to estimate
parameters of both categories jointly. Likewise, the use of these methods for
hydrogeological parameter using multiple injection (or leak) or abstraction events (i.e.
as in hydraulic tomography or tracer tomography) (e.g. Tso et al., 2016; Zha et al., 2019)
have potential to improve characterisation of environmental sites.
Bibliography
Aalstad, K., Westermann, S., Schuler, T.V., Boike, J., Bertino, L., 2018. Ensemble-based
assimilation of fractional snow-covered area satellite retrievals to estimate the
snow distribution at Arctic sites. Cryosphere 12, 247–270.
https://doi.org/10.5194/tc-12-247-2018
Abdulsamad, F., Revil, A., Soueid Ahmed, A., Coperey, A., Karaoulis, M., Nicaise, S.,
Bibliography
258
Peyras, L., 2019. Induced polarization tomography applied to the detection and
the monitoring of leaks in embankments. Eng. Geol. 254, 89–101.
https://doi.org/10.1016/J.ENGGEO.2019.04.001
Aghasi, A., Mendoza-Sanchez, I., Miller, E.L., Ramsburg, C.A., Abriola, L.M., 2013. A
geometric approach to joint inversion with applications to contaminant source
zone characterization. Inverse Probl. 29, 115014. https://doi.org/10.1088/0266-
5611/29/11/115014
Ahlbom, K., Carlsson, L., Olsson, O., 1983. Final disposal of spent nuclear fuel -
geological, hydrogeological and geophysical methods for site characterization.
Stockholm.
Ahlstrom, S.W., Foote, H.P., Arnett, R.C., Cole, C.R., Serne, R.J., 1977. Multicomponent
mass transport model: theory and numerical implementation (discrete-parcel-
random-walk version). Richland, Washington.
Ahmed, M.A., Passioura, J., Carminati, A., 2018. Hydraulic processes in roots and the
rhizosphere pertinent to increasing yield of water-limited grain crops: a critical
review. J. Exp. Bot. 69, 3255–3265. https://doi.org/10.1093/jxb/ery183
Alpak, F., Torres-Verdin, C., Habashy, T.M., 2004. Joint inversion of transient pressure
and dc resistivity measurements acquired with in-situ permanent sensors: A
numerical study. Geophysics 69, 1173–1191. https://doi.org/10.1190/1.1801935
Altdorff, D., von Hebel, C., Borchard, N., van der Kruk, J., Bogena, H.R., Vereecken,
H., Huisman, J.A., 2017. Potential of catchment-wide soil water content prediction
using electromagnetic induction in a forest ecosystem. Environ. Earth Sci. 76, 111.
https://doi.org/10.1007/s12665-016-6361-3
Alumbaugh, D.L., Newman, G.A., 2000. Image appraisal for 2-D and 3-D
electromagnetic inversion. Geophysics 65, 1455. https://doi.org/10.1190/1.1444834
Amidu, S.A., Dunbar, J.A., 2007. Geoelectric Studies of Seasonal Wetting and Drying
of a Texas Vertisol. Vadose Zo. J. 6, 511. https://doi.org/10.2136/vzj2007.0005
Andersen, K.E., Brooks, S.P., Hansen, M.B., 2003. Bayesian inversion of geoelectrical
resistivity data. J. R. Stat. Soc. Ser. B Stat. Methodol. 65, 619–642.
https://doi.org/10.1111/1467-9868.00406
Artiola, J.F., Pepper, I.L., Brusseau, M.L., 2004. Environmental Monitoring and
Characterization. Elsevier. https://doi.org/10.1016/B978-0-12-064477-3.X5000-0
Asch, T., Morrison, H.F., 1989. Mapping and monitoring electrical resistivity with
surface and subsurface electrode arrays. Geophysics 54, 235–244.
Asher, M.J.J., Croke, B.F.W.F.W., Jakeman, A.J.J., Peeters, L.J.M.J.M., 2015. A review of
surrogate models and their application to groundwater modeling. Water Resour.
Res. 51, 5957–5973. https://doi.org/10.1002/2015WR016967
Aster, R., Borchers, B., Thurber, C.H., 2005. Parameter estimation and inverse problems.
Bibliography
259
Elsevier, Burlington, Massachusetts, USA.
Audebert, M., Clément, R., Touze-Foltz, N., Günther, T., Moreau, S., Duquennoi, C.,
2014. Time-lapse ERT interpretation methodology for leachate injection
monitoring based on multiple inversions and a clustering strategy (MICS). J. Appl.
Geophys. 111, 320–333. https://doi.org/10.1016/j.jappgeo.2014.09.024
Back, P.-E., 2007. A model for estimating the value of sampling programs and the
optimal number of samples for contaminated soil. Environ. Geol. 52, 573–585.
https://doi.org/10.1007/s00254-006-0488-6
Backus, G.E., Gilbert, F., 1970. Uniqueness in the Inversion of inaccurate Gross Earth
Data. Philos. Trans. R. Soc. London A 266, 123–192.
Backus, G.E., Gilbert, F., 1968. The Resolving power of Gross Earth Data. Geophys. J.
R. Astron. Soc. 16, 169–205.
Balbarini, N., Rønde, V., Maurya, P., Fiandaca, G., Møller, I., Erik Klint, K.,
Christiansen, A. V., Binning, P.J., Bjerg, P.L., 2018. Geophysics Based Contaminant
Mass Discharge Quantification Downgradient of a Landfill and a Former
Pharmaceutical Factory. Water Resour. Res.
https://doi.org/10.1029/2017WR021855
Barker, J.A., 1991. The reciprocity principle and an analytical solution for Darcian flow
in a network. Water Resour. Res. 27, 743–746. https://doi.org/10.1029/91WR00258
Barnett, D.B., Gee, G.W., Sweeney, M.D., 2002. Results of Tank-Leak Detection
Demonstration Using Geophysical Techniques at the Hanford Mock Tank Site—
Fiscal Year 2001. Richland, Washington.
Barnett, D.B., Gee, G.W., Sweeney, M.D., Johnson, M.D., Medina, V.F., Khan, F., 2003.
Results of Performance Evaluation Testing of Electrical Leak-Detection Methods
at the Hanford Site Mock Tank – FY 2002-2003. Richland, Washington.
Bass, B., Cardenas, M.B., Befus, K.M., 2017. Seasonal Shifts in Soil Moisture throughout
a Semiarid Hillslope Ecotone during Drought: A Geoelectrical View. Vadose Zo.
J. 0, 0. https://doi.org/10.2136/vzj2016.11.0108
Bates, D., Maechler, M., Bolker, B.M., Walker, S., 2015. Fitting Linear Mixed-Effects
Models using lme4. J. Stat. Softw. 67, 1–48. https://doi.org/10.18637/jss.v067.i01
Bayliss, C.R., Langley, K.F., 2003. Nuclear Decommissioning, Waste Management, and
Environmental Site Remediation, Nuclear Decommissioning, Waste Management,
and Environmental Site Remediation. Elsevier. https://doi.org/10.1016/B978-
075067744-8/50027-8
Bazin, S., Pfaffhuber, A.A., 2013. Mapping of quick clay by electrical resistivity
tomography under structural constraint. J. Appl. Geophys. 98, 280–287.
https://doi.org/10.1016/J.JAPPGEO.2013.09.002
Beff, L., Günther, T., Vandoorne, B., Couvreur, V., Javaux, M., 2013. Three-dimensional
Bibliography
260
monitoring of soil water content in a maize field using Electrical Resistivity
Tomography. Hydrol. Earth Syst. Sci. 17, 595–609. https://doi.org/10.5194/hess-17-
595-2013
Begg, S., Bratvold, R., Campbell, J., 2013. The value of flexibility in managing
Uncertainty in oil and gas investments, in: SPE Annual Technical Conference and
Exhibition. Society of Petroleum Engineers. https://doi.org/10.2118/77586-MS
Beskardes, G.D., Weiss, C.J., 2018. Modelling dc responses of 3-D complex fracture
networks. Geophys. J. Int. 214, 1901–1912. https://doi.org/10.1093/gji/ggy234
Beven, K., 2006. A manifesto for the equifinality thesis. J. Hydrol. 320, 18–36.
https://doi.org/10.1016/j.jhydrol.2005.07.007
Beven, K., Westerberg, I., 2011. On red herrings and real herrings: disinformation and
information in hydrological inference. Hydrol. Process. 25, 1676–1680.
https://doi.org/10.1002/hyp.7963
Bickel, E.J., Bratvold, R.B., 2008. From uncertainty quantification to decision making in
the oil and gas industry. Energy, Explor. Exploit. 26, 311–325.
https://doi.org/10.1260/014459808787945344
Binley, A., 2015a. Tools and Techniques: Electrical Methods, Treatise on Geophysics.
Elsevier B.V. https://doi.org/10.1016/B978-0-444-53802-4.00192-5
Binley, A., 2015b. Tools and techniques: Electrical methods. Treatise Geophys.
https://doi.org/10.1016/B978-0-444-53802-4.00192-5
Binley, A., 2003. Vadose Zone Tracer Testing in the UK Sherwood Sandstone :
Hydrogeophysical Data Report. Livermore, California.
Binley, A., Beven, K., 2003. Vadose zone flow model uncertianty as conditioned on
geophysical data. Groundwater 41, 119–127.
Binley, A., Cassiani, G., Middleton, R., Winship, P., 2002a. Vadose zone flow model
parameterisation using cross-borehole radar and resistivity imaging. J. Hydrol.
267, 147–159. https://doi.org/10.1016/S0022-1694(02)00146-4
Binley, A., Cassiani, G., Winship, P., 2004. Characterization of heterogeneity in
unsaturated sandstone using borehole logs and cross-borehole tomography, in:
Aquifer Characterization, in: Aquifer Characterization. SEPM (Society for
Sedimentary Geology), pp. 129–138. https://doi.org/10.2110/pec.04.80.0129
Binley, A., Daily, W., Ramirez, A., 1997. Detecting leaks from environmental barriers
using electrical current imaging. J. Environ. Eng. Geophys. 2 (March), 11–19.
https://doi.org/10.4133/JEEG2.1.11
Binley, A., Hubbard, S.S., Huisman, J.A., Revil, A., Robinson, D.A., Singha, K., Slater,
L.D., 2015. The emergence of hydrogeophsyics for improved understanding of
subsurface processes over multiple scales. Water Resour. Res. 51, 3837–3866.
https://doi.org/10.1002/2015WR017016
Bibliography
261
Binley, A., Ramirez, A.L., Daily, W., 1995. Regularised image reconstruction of noisy
electrical resistance data, in: Proceedings of the 4th Workshop of the European
Concerted Action on Process Tomography,. Bergen, pp. 401–410.
Binley, A., Slater, L.D., Fukes, M., Cassiani, G., 2005. Relationship between spectral
induced polarization and hydraulic properties of saturated and unsaturated
sandstone. Water Resour. Res. 41, W12417. https://doi.org/10.1029/2005WR004202
Binley, A., Winship, P., Middleton, R., Pokar, M., West, J., 2001. High-resolution
characterization of vadose zone dynamics using cross-borehole radar. Water
Resour. Res. 37, 2639–2652. https://doi.org/10.1029/2000WR000089
Binley, A., Winship, P., West, L.J., Pokar, M., Middleton, R., 2002b. Seasonal variation
of moisture content in unsaturated sandstone inferred from borehole radar and
resistivity profiles. J. Hydrol. 267, 160–172. https://doi.org/10.1016/S0022-
1694(02)00147-6
Blair, G.S., Henrys, P., Leeson, A., Watkins, J., Eastoe, E., Jarvis, S., Young, P.J., 2019.
Data Science of the Natural Environment: A Research Roadmap. Front. Environ.
Sci. 7. https://doi.org/10.3389/fenvs.2019.00121
Blair, G.S., Watkins, J., August, T., Bassett, R., Brown, M., Ciar, D., Dean, G., Henrys,
P., Edwards, L., Nundloll, V., Samreen, F., Simm, W., Towe, R., 2018. Virtual Data
Labs: Technological Support for Complex, Collaborative Research in the
Environmental Sciences, in: CMWR Conference (Computational Methods in
Water Resources). St Malo, France.
Blome, M., Maurer, H., Greenhalgh, S., 2011. Geoelectric experimental design —
Efficient acquisition and exploitation of complete pole-bipole data sets.
GEOPHYSICS 76, F15–F26. https://doi.org/10.1190/1.3511350
Booth, P.M., 1997. Sellafield site characterisation project: an appraisal of site
investigation techniques previously untrailed at BNFL’s sites. Risley, United
Kingdom.
Bouchedda, A., Chouteau, M., Binley, A., Giroux, B., 2012. 2-D joint structural
inversion of cross-hole electrical resistance and ground penetrating radar data. J.
Appl. Geophys. 78, 52–67. https://doi.org/10.1016/j.jappgeo.2011.10.009
Bouchedda, A., Giroux, B., Gloaguen, E., 2015. Bayesian ERT Inversion using Non-
Stationary Inverse Matérn Covariance Matrix, in: SEG Annual Meeting. New
Orleans.
Bowden, R.A., Bumpus, C., Littleboy, A.K., 1998. An overview and update of the site
characterization studies at Sellafield. Proc. Yorksh. Geol. Soc. 52, 125–137.
https://doi.org/10.1144/pygs.52.2.125
Bratvold, R.B., Begg, S., 2010. Making good decisions. Society of Petroleum Engineers,
Richardson, Texas, USA.
Brillante, L., Bois, B., Mathieu, O., Bichet, V., Michot, D., Lévêque, J., 2014. Monitoring
Bibliography
262
soil volume wetness in heterogeneous soils by electrical resistivity. A field-based
pedotransfer function. J. Hydrol. 516, 56–66.
https://doi.org/10.1016/j.jhydrol.2014.01.052
Brillante, L., Mathieu, O., Bois, B., Van Leeuwen, C., Lévêque, J., 2015. The use of soil
electrical resistivity to monitor plant and soil water relationships in vineyards.
Soil 1, 273–286. https://doi.org/10.5194/soil-1-273-2015
Bruggeman, G.A., 1972. The reciprocity principle in flow through heterogeneous
porous media, in: Fundamentals of Transport Phenomena in Porous Media,.
Elsevier, New York, p. 136.
Brunet, P., Clément, R., Bouvier, C., 2010. Monitoring soil water content and deficit
using Electrical Resistivity Tomography (ERT) - A case study in the Cevennes area,
France. J. Hydrol. 380, 146–153. https://doi.org/10.1016/j.jhydrol.2009.10.032
Brunetti, C., 2019. Bayesian model selection in hydrogeophysics and hydrogeology.
Université de Lausanne.
Brunetti, C., Linde, N., 2018. Impact of petrophysical uncertainty on Bayesian
hydrogeophysical inversion and model selection. Adv. Water Resour. 111, 346–
359. https://doi.org/10.1016/j.advwatres.2017.11.028
Brusseau, M.L., Guo, Z., 2014. Assessing contaminant-removal conditions and plume
persistence through analysis of data from long-term pump-and-treat operations.
J. Contam. Hydrol. 164, 16–24. https://doi.org/10.1016/j.jconhyd.2014.05.004
Cai, Z., Wilson, R.D., Cardiff, M.A., Kitanidis, P.K., 2011. Increasing Confidence in
Mass Discharge Estimates Using Geostatistical Methods. Ground Water 49, 197–
208. https://doi.org/10.1111/j.1745-6584.2010.00709.x
Calendine, S., Rucker, D.F., Fink, J.B., Levitt, M.T., 2011. Automated leak detection of
buried tanks using geophysical methods at the Hanford nuclear site, in: SAGEEP
2011. Annual Meeting of the Environmental and Engineering Geophysical Society.
Charleston, South Carolina.
Camporese, M., Cassiani, G., Deiana, R., Salandin, P., 2011. Assessment of local
hydraulic properties from electrical resistivity tomography monitoring of a three-
dimensional synthetic tracer test experiment. Water Resour. Res. 47, 1–15.
https://doi.org/10.1029/2011WR010528
Camporese, M., Cassiani, G., Deiana, R., Salandin, P., Binley, A., 2015. Coupled and
uncoupled hydrogeophysical inversions using ensemble Kalman filter
assimilation of ERT-monitored tracer test data. Water Resour. Res. 51, 3277–3291.
https://doi.org/10.1002/2014WR016017
Carey, A.M., Paige, G.B., Carr, B.J., Dogan, M., 2017. Forward modeling to investigate
inversion artifacts resulting from time-lapse electrical resistivity tomography
during rainfall simulations. J. Appl. Geophys. 145, 39–49.
https://doi.org/10.1016/J.JAPPGEO.2017.08.002
Bibliography
263
Carrassi, A., Bocquet, M., Bertino, L., Evensen, G., 2018. Data assimilation in the
geosciences: An overview of methods, issues, and perspectives. Wiley Interdiscip.
Rev. Clim. Chang. 9, e535. https://doi.org/10.1002/wcc.535
Carrigan, C.R., Yang, X., LaBrecque, D.J., Larsen, D., Freeman, D., Ramirez, A.L., Daily,
W., Aines, R., Newmark, R., Friedmann, J., Hovorka, S., 2013. Electrical resistance
tomographic monitoring of CO2 movement in deep geologic reservoirs. Int. J.
Greenh. Gas Control 18, 401–408. https://doi.org/10.1016/j.ijggc.2013.04.016
Cassiani, G., Binley, A., 2005. Modeling unsaturated flow in a layered formation under
quasi-steady state conditions using geophysical data constraints. Adv. Water
Resour. 28, 467–477. https://doi.org/10.1016/j.advwatres.2004.12.007
Cassiani, G., Bruno, V., Villa, A., Fusi, N., Binley, A.M., 2006. A saline tracer test
monitored via time-lapse surface electrical resistivity tomography. J. Appl.
Geophys. 59, 244–259. https://doi.org/10.1016/j.jappgeo.2005.10.007
Cassiani, G., Godio, A., Stocco, S., Villa, A., Deiana, R., Frattini, P., Rossi, M., 2009.
Monitoring the hydrologic behaviour of a mountain slope via time-lapse electrical
resistivity tomography. Near Surf. Geophys. 7, 475–486.
https://doi.org/10.3997/1873-0604.2009013
Caterina, D., Beaujean, J., Robert, T., Nguyen, F., 2013. A comparison study of different
image appraisal tools for electrical resistivity tomography. Near Surf. Geophys.
11, 639–657. https://doi.org/10.3997/1873-0604.2013022
Caterina, D., Hermans, T., Nguyen, F., 2014. Case studies of incorporation of prior
information in electrical resistivity tomography: comparison of different
approaches. Near Surf. Geophys. 12, 451–465. https://doi.org/10.3997/1873-
0604.2013070
Chada, N.K., Iglesias, M.A., Roininen, L., Stuart, A.M., 2018. Parameterizations for
ensemble Kalman inversion. https://doi.org/10.1088/1361-6420/aab6d9
Chambers, J.E., Gunn, D.A., Wilkinson, P.B., Meldrum, P.I., Haslam, E., Holyoake, S.,
Kirkham, M., Kuras, O., Merritt, A., Wragg, J., 2014a. 4D electrical resistivity
tomography monitoring of soil moisture dynamics in an operational railway
embankment. Near Surf. Geophys. 12, 61–72. https://doi.org/10.3997/1873-
0604.2013002
Chambers, J.E., Kuras, O., Meldrum, P.I., Ogilvy, R.D., Hollands, J., 2006. Electrical
resistivity tomography applied to geologic, hydrogeologic, and engineering
investigations at a former waste-disposal site. Geophysics 71, B231–B239.
https://doi.org/10.1190/1.2360184
Chambers, J.E., Wilkinson, P.B., Penn, S., Meldrum, P.I., Kuras, O., Loke, M.H., Gunn,
D.A., 2013. River terrace sand and gravel deposit reserve estimation using three-
dimensional electrical resistivity tomography for bedrock surface detection. J.
Appl. Geophys. 93, 25–32. https://doi.org/10.1016/j.jappgeo.2013.03.002
Bibliography
264
Chambers, J.E., Wilkinson, P.B., Uhlemann, S., Sorensen, J.P.R., Roberts, C., Newell,
A.J., Ward, W.O.C., Binley, A., Williams, P.J., Gooddy, D.C., Old, G., Bai, L., 2014b.
Derivation of lowland riparian wetland deposit architecture using geophysical
image analysis and interface detection. Water Resour. Res. 50, 5886–5905.
https://doi.org/10.1002/2014WR015643
Chambers, J.E., Wilkinson, P.B., Wardrop, D., Hameed, A., Hill, I., Jeffrey, C., Loke,
M.H., Meldrum, P.I., Kuras, O., Cave, M., Gunn, D.A., 2012. Bedrock detection
beneath river terrace deposits using three-dimensional electrical resistivity
tomography. Geomorphology 177–178, 17–25.
https://doi.org/10.1016/j.geomorph.2012.03.034
Chen, H.C., Fang, J.H., 1986. Sensitivity analysis of the parameters in Archie’s water
saturation equation. Log Anal. 27, 39–44. https://doi.org/SPWLA-1986-
vXXVIIn5a3
Chen, J., Hubbard, S., Peterson, J., Williams, K., Fienen, M., Jardine, P., Watson, D.,
2006. Development of a joint hydrogeophysical inversion approach and
application to a contaminated fractured aquifer. Water Resour. Res. 42, 1–13.
https://doi.org/10.1029/2005WR004694
Chen, J., Hubbard, S.S., Williams, K.H., 2013. Data-driven approach to identify field-
scale biogeochemical transitions using geochemical and geophysical data and
hidden Markov models: Development and application at a uranium-
contaminated aquifer. Water Resour. Res. 49, 6412–6424.
https://doi.org/10.1002/wrcr.20524
Chen, J., Hubbard, S.S., Williams, K.H., Ficklin, D.L., 2016. Estimating groundwater
dynamics at a Colorado River floodplain site using historical hydrological data
and climate information. Water Resour. Res. n/a-n/a.
https://doi.org/10.1002/2015WR017777
Chen, J.S., Hubbard, S.S., Gaines, D., Korneev, V., Baker, G., Watson, D., 2010.
Stochastic estimation of aquifer geometry using seismic refraction data with
borehole depth constraints. Water Resour. Res. 46, 16. https://doi.org/W11539
10.1029/2009wr008715
Chen, X., Hammond, G.E., Murray, C.J., Rockhold, M.L., Vermeul, V.R., Zachara, J.M.,
2013. Application of ensemble-based data assimilation techniques for aquifer
characterization using tracer data at Hanford 300 area. Water Resour. Res. 49,
7064–7076. https://doi.org/10.1002/2012WR013285
Chen, X., Murakami, H., Hahn, M.S., Hammond, G.E., Rockhold, M.L., Zachara, J.M.,
Rubin, Y., 2012. Three-dimensional Bayesian geostatistical aquifer
characterization at the Hanford 300 Area using tracer test data. Water Resour. Res.
48, W06501. https://doi.org/10.1029/2011WR010675
Chen, Z., Gómez-Hernández, J.J., Xu, T., Zanini, A., 2018. Joint identification of
contaminant source and aquifer geometry in a sandbox experiment with the
restart ensemble Kalman filter. J. Hydrol. 564, 1074–1084.
Bibliography
265
https://doi.org/10.1016/J.JHYDROL.2018.07.073
Christ, J.A., Ramsburg, C.A., Pennell, K.D., Abriola, L.M., 2010. Predicting DNAPL
mass discharge from pool-dominated source zones. J. Contam. Hydrol.
https://doi.org/10.1016/j.jconhyd.2010.02.005
Christ, J.A., Ramsburg, C.A., Pennell, K.D., Abriola, L.M., 2006. Estimating mass
discharge from dense nonaqueous phase liquid source zones using upscaled mass
transfer coefficients: An evaluation using multiphase numerical simulations.
Water Resour. Res. 42, 1–13. https://doi.org/10.1029/2006WR004886
Christakos, G., 2000. Modern Spatiotemporal Geostatistics. Oxford University Press,
New York.
Christensen, S., Doherty, J., 2008. Predictive error dependencies when using pilot
points and singular value decomposition in groundwater model calibration. Adv.
Water Resour. 31, 674–700. https://doi.org/10.1016/J.ADVWATRES.2008.01.003
Cooper, S., 1997. Assessment of the potential for downhole geophysics within the BHP
boreholes. Risley, United Kingdom.
Cooper, S., Ross, M.O., 2005. Application and comparative assessment of non-intrusive
site characterisation techniques. Daresbury, United Kingdom.
Coscia, I., Greenhalgh, S.A., Linde, N., Doetsch, J., Marescot, L., Gunther, T., Vogt, T.,
Green, A.G., 2011. 3D crosshole ERT for aquifer characterization and monitoring
of infiltrating river water. Geophysics 76, G49–G59.
https://doi.org/10.1190/1.3553003
Crestani, E., 2013. Tracer Test Assimilation Data for the Assessment of Local Hydraulic
Properties in Heterogeneous Aquifers. University of Padua.
Crestani, E., Camporese, M., Salandin, P., 2015. Assessment of hydraulic conductivity
distributions through assimilation of travel time data from ERT-monitored tracer
tests. Adv. Water Resour. 84, 23–36.
https://doi.org/10.1016/j.advwatres.2015.07.022
Curtis, A., Lomax, A., 2001. Prior information, sampling distributions, and the curse of
dimensionality. GEOPHYSICS 66, 372–378. https://doi.org/10.1190/1.1444928
Curtis, A., Spencer, C., 1999. Survey design strategies for linearized nonlinear
inversion, in: SEG Technical Program Expanded Abstracts 1999. Society of
Exploration Geophysicists, pp. 1775–1778. https://doi.org/10.1190/1.1820882
Dahlin, T., 2000. Short note on electrode charge-up effects in DC resistivity data
acquistiion using multi-electrode arrays. Geophys. Prospect. 48, 181–187.
https://doi.org/10.1046/j.1365-2478.2000.00172.x
Dahlin, T., Dalsegg, E., Sandstrom, T., 2013. Data quality quantification for time
domain IP data acquired along a planned tunnel near Oslo, Norway, in: Near
Surface Geoscience 2013--19th European Meeting of Environmental and
Bibliography
266
Engineering Geophysics. Bochum, Germany.
Daily, W., Ramirez, A., 1995. Electrical resistance tomography during in-situ
trichloroethylene remediation at the Savannah River Site. J. Appl. Geophys. 33,
239–249. https://doi.org/10.1016/0926-9851(95)90044-6
Daily, W., Ramirez, A., Binley, A., 2004. Remote monitoring of leaks in storage tanks
using electrical resistance tomography: Application at the Hanford Site. J. Environ.
Eng. Geophys. 9, 11–24. https://doi.org/10.4133/jeeg9.1.11
Daily, W., Ramirez, A., Binley, A., LaBrecque, D., 2005. 17. Electrical resistance
tomography—theory and practice, in: Butler, D.K. (Ed.), Near-Surface Geophysics.
Society of Exploration Geophysicists, Tulsa, Oklahoma, USA, pp. 525–550.
https://doi.org/10.1190/1.9781560801719.ch17
Daily, W., Ramirez, A., LaBrecque, D., Nitao, J., 1992. Electrical resistivity tomography
of vadose water movement. Water Resour. Res. 28, 1429–1442.
https://doi.org/10.1029/91WR03087
Datta, B., Chakrabarty, D., Dhar, A., 2009. Simultaneous identification of unknown
groundwater pollution sources and estimation of aquifer parameters. J. Hydrol.
376, 48–57. https://doi.org/10.1016/J.JHYDROL.2009.07.014
Day-Lewis, F.D., Johnson, C.D., Singha, K., Lane Jr, J.W., 2008. Best practices in
electrical resistivity imaging: Data collection and processing, and application to
data from Corinna, Maine. EPA report, Boston, MA.
Day-Lewis, F.D., Lane, J.W., 2004. Assessing the resolution-dependent utility of
tomograms for geostatistics. Geophys. Res. Lett. 31, 2–5.
https://doi.org/10.1029/2004GL019617
Day-Lewis, F.D., Singha, K., Binley, A.M., 2005. Applying petrophysical models to
radar travel time and electrical resistivity tomograms: Resolution-dependent
limitations. J. Geophys. Res. B Solid Earth 110, 1–17.
https://doi.org/10.1029/2004JB003569
de Barros, F.P.J., Ezzedine, S., Rubin, Y., 2012. Impact of hydrogeological data on
measures of uncertainty, site characterization and environmental performance
metrics. Adv. Water Resour. 36, 51–63.
https://doi.org/10.1016/j.advwatres.2011.05.004
de Barros, F.P.J., Rubin, Y., 2008. A risk-driven approach for subsurface site
characterization. Water Resour. Res. 44, 1–14.
https://doi.org/10.1029/2007WR006081
de Pasquale, G., Linde, N., Doetsch, J., Holbrook, W.S., 2019. Probabilistic inference of
subsurface heterogeneity and interface geometry using geophysical data.
Geophys. J. Int. https://doi.org/10.1093/gji/ggz055
Deceuster, J., Kaufmann, O., Camp, M. Van, 2013. Automated identification of changes
in electrode contact properties for long-term permanent ERT monitoring
Bibliography
267
experiments. Geophysics 78, E79–E94. https://doi.org/10.1190/GEO2012-0088.1
Delay, F., Ackerer, P., Guadagnini, A., 2011. Theoretical analysis and field evidence of
reciprocity gaps during interference pumping tests. Adv. Water Resour. 34, 592–
606. https://doi.org/10.1016/j.advwatres.2011.02.006
Deman, G., Konakli, K., Sudret, B., Kerrou, J., Perrochet, P., Benabderrahmane, H., 2016.
Using sparse polynomial chaos expansions for the global sensitivity analysis of
groundwater lifetime expectancy in a multi-layered hydrogeological model.
Reliab. Eng. Syst. Saf. 147, 156–169. https://doi.org/10.1016/j.ress.2015.11.005
Demirel, S., Irving, J., Roubinet, D., 2019. Comparison of REV size and tensor
characteristics for the electrical and hydraulic conductivities in fractured rock.
Geophys. J. Int. 216, 1953–1973. https://doi.org/10.1093/gji/ggy537
Diggle, P.J., Heagerty, P., Liang, K.-Y., Zeger, S., 2015. Analysis of Longitudinal Data,
2nd ed. Oxford University Press.
Doetsch, J., 2011. Joint and constrained inversion of geophsyical data for improved
imaging of aquier structure and processes. ETH Zurich.
Doetsch, J., Coscia, I., Greenhalgh, S., Linde, N., Green, A., Gunther, T., 2010a. The
borehole-fluid effect in electrical resistivity imaging. Geophysics 75, F107–F114.
https://doi.org/Doi 10.1190/1.3467824
Doetsch, J., Kowalsky, M.B., Doughty, C., Finsterle, S., Ajo-Franklin, J.B., Carrigan, C.R.,
Yang, X., Hovorka, S.D., Daley, T.M., 2013. Constraining CO2 simulations by
coupled modeling and inversion of electrical resistance and gas composition data.
Int. J. Greenh. Gas Control 18, 510–522. https://doi.org/10.1016/j.ijggc.2013.04.011
Doetsch, J., Linde, N., Binley, A., 2010b. Structural joint inversion of time-lapse
crosshole ERT and GPR traveltime data. Geophys. Res. Lett. 37, 1–6.
https://doi.org/10.1029/2010GL045482
Doetsch, J., Linde, N., Pessognelli, M., Green, A.G., Gunther, T., 2012a. Constraining 3-
D electrical resistance tomography with GPR reflection data for improved aquifer
characterization. J. Appl. Geophys. 78, 68–76.
https://doi.org/10.1016/j.jappgeo.2011.04.008
Doetsch, J., Linde, N., Vogt, T., Binley, A., Green, A.G., 2012b. Imaging and quantifying
salt-tracer transport in a riparian groundwater system by means of 3D ERT
monitoring. Geophysics 77, B207. https://doi.org/10.1190/geo2012-0046.1
Dumont, G., Pilawski, T., Dzaomuho-Lenieregue, P., Hiligsmann, S., Delvigne, F.,
Thonart, P., Robert, T., Nguyen, F., Hermans, T., 2016. Gravimetric water
distribution assessment from geoelectrical methods (ERT and EMI) in municipal
solid waste landfill. Waste Manag. 55, 129–140.
https://doi.org/10.1016/J.WASMAN.2016.02.013
Dumont, G., Pilawski, T., Hermans, T., Nguyen, F., Garré, S., 2018. The effect of initial
water distribution and spatial resolution on the interpretation of ERT monitoring
Bibliography
268
of water infiltration in a landfill cover. Hydrol. Earth Syst. Sci. Discuss. 1–26.
https://doi.org/10.5194/hess-2018-163
Eddy-Dilek, C.A., Looney, B.B., Hoekstra, P., Harthill, N., Blohm, M., Phillips, D.B.,
1997. Definition of a critical confining zone using surface geophysical methods.
Groundwater 35, 451–462.
Edwards, L.S., 1977. A modified pseudosection for resistivity and IP. Geophysics 42,
1020–1036. https://doi.org/10.1190/1.1440762
Eidsvik, J., Mukerji, T., Bhattacharjya, D., 2015. Value of information in the earth
sciences: Integrating spatial modeling and decision analysis. Cambridge
University Press, Cambridge.
Elwaseif, M., Slater, L., 2012. Improved Resistivity Imaging of Targets with Sharp
Boundaries Using an Iterative Disconnect Procedure. J. Environ. Eng. Geophys.
17, 89–101. https://doi.org/10.2113/JEEG17.2.89
Emerick, A.A., Reynolds, A.C., 2013. Ensemble smoother with multiple data
assimilation. Comput. Geosci. 55, 3–15. https://doi.org/10.1016/j.cageo.2012.03.011
Enescu, N., Ahokas, T., Heikkinen, E., Cosma, C., 2004. Integrated Geophysical
Characterization of a Hard Rock Site, in: EAGE 66th Conference & Exhibition.
Paris, France, pp. 7–10.
Environment Agency, 2011. Geophysical surveying techniques to characterise a site for
a geological disposal facility: A review of recent developments and NDA’s
proposals.
Eslick, J.C., Ng, B., Gao, Q., Tong, C.H., Sahinidis, N. V., Miller, D.C., 2014. A
framework for optimization and quantification of uncertainty and sensitivity for
developing carbon capture systems. Energy Procedia 63, 1055–1063.
https://doi.org/10.1016/j.egypro.2014.11.113
Evensen, G., 2018. Analysis of iterative ensemble smoothers for solving inverse
problems. Comput. Geosci. 22, 885–908. https://doi.org/10.1007/s10596-018-9731-
y
Falade, G.K., 1981. Analysis of the reciprocity concept in a porous medium. Water
Resour. Res. 17, 918–920. https://doi.org/10.1029/WR017i004p00918
Falzone, S., Robinson, J., Slater, L., 2018. Characterization and Monitoring of Porous
Media with Electrical Imaging: A Review. Transp. Porous Media 1–26.
https://doi.org/10.1007/s11242-018-1203-2
Fan, J., Scheuermann, A., Guyot, A., Baumgartl, T., Lockington, D.A., 2015.
Quantifying spatiotemporal dynamics of root-zone soil water in a mixed forest on
subtropical coastal sand dune using surface ERT and spatial TDR. J. Hydrol. 523,
475–488. https://doi.org/10.1016/j.jhydrol.2015.01.064
Farquharson, C.G., 2008. Constructing piecewise-constant models in multidimensional
Bibliography
269
minimum-structure inversions. GEOPHYSICS 73, K1–K9.
https://doi.org/10.1190/1.2816650
Fernández-Martínez, J.L., Fernández-Muñiz, Z., Xu, S., Cernea, A., Sirieix, C., Riss, J.,
2019. Efficient uncertainty analysis of the 3D electrical tomography inverse
problem. GEOPHYSICS 1–62. https://doi.org/10.1190/geo2017-0729.1
Fernández-Muñiz, Z., Khaniani, H., Fernández-Martínez, J.L., 2019. Data kit inversion
and uncertainty analysis. J. Appl. Geophys. 161, 228–238.
https://doi.org/10.1016/j.jappgeo.2018.12.022
Feyen, L., Gorelick, S.M., 2005. Framework to evaluate the worth of hydraulic
conductivity data for optimal groundwater resources management in ecologically
sensitive areas. Water Resour. Res. 41, 1–13.
https://doi.org/10.1029/2003WR002901
Fiandaca, G., Doetsch, J., Vignoli, G., Auken, E., 2015. Generalized focusing of time-
lapse changes with applications to direct current and time-domain induced
polarization inversions 1101–1112. https://doi.org/10.1093/gji/ggv350
Flores-Orozco, A., Kemna, A., Zimmermann, E., 2012. Data error quantification in
spectral induced polarization imaging. Geophysics 77, E227.
https://doi.org/10.1190/geo2010-0194.1
Flores-Orozco, A., Williams, K.H., Kemna, A., 2013. Time-lapse spectral induced
polarization imaging of stimulated uranium bioremediation. Near Surf. Geophys.
11, 531–544. https://doi.org/10.3997/1873-0604.2013020
Flores-Orozco, A., Williams, K.H., Long, P.E., Hubbard, S.S., Kemna, A., 2011. Using
complex resistivity imaging to infer biogeochemical processes associated with
bioremediation of an uranium-contaminated aquifer. J. Geophys. Res.
Biogeosciences 116, 1–17. https://doi.org/10.1029/2010JG001591
Frangos, W., Ter-Saakian, S., 1996. Resistivity and induced polarization survey at a
Russian nuclear waste site. Berkeley, California, United States.
https://doi.org/10.1149/1.2221597
French, H.K., Kästner, M., van der Zee, S.E.A.T.M., 2014. New approaches for low-
invasive contaminated site characterization, monitoring and modelling. Environ.
Sci. Pollut. Res. 21, 8893–8896. https://doi.org/10.1007/s11356-014-2840-9
Friedel, S., 2003. Resolution, stability and efficiency of resistivity tomography
estimated from a generalized inverse approach. Geophys. J. Int. 153, 305–316.
https://doi.org/10.1046/j.1365-246X.2003.01890.x
Friedman, S.P., 2005. Soil properties influencing apparent electrical conductivity: a
review. Comput. Electron. Agric. 46, 45–70.
https://doi.org/10.1016/j.compag.2004.11.001
Fu, J., Jaime Gómez-Hernández, J., 2009. Uncertainty assessment and data worth in
groundwater flow and mass transport modeling using a blocking Markov chain
Bibliography
270
Monte Carlo method. J. Hydrol. 364, 328–341.
https://doi.org/10.1016/j.jhydrol.2008.11.014
Furman, A., Ferré, T.P.A., Heath, G.L., 2007. Spatial focusing of electrical resistivity
surveys considering geologic and hydrologic layering. Geophysics 72, F65.
https://doi.org/10.1190/1.2433737
Furman, A., Ferré, T.P.A., Warrick, A.W., 2004. Optimization of ERT Surveys for
Monitoring Transient Hydrological Events Using Perturbation Sensitivity and
Genetic Algorithms. Vadose Zo. J. 3, 1230. https://doi.org/10.2136/vzj2004.1230
Galetti, E., Curtis, A., 2018. Transdimensional Electrical Resistivity Tomography. J.
Geophys. Res. Solid Earth. https://doi.org/10.1029/2017JB015418
Gallardo, L.A., Meju, M.A., 2011. Structure-coupled multiphysics imaging in
geophysical sciences. Rev. Geophys. 49, RG1003.
https://doi.org/10.1029/2010RG000330
Gan, Y., Duan, Q., Gong, W., Tong, C., Sun, Y., Chu, W., Ye, A., Miao, C., Di, Z., 2014.
A comprehensive evaluation of various sensitivity analysis methods: A case study
with a hydrological model. Environ. Model. Softw. 51, 269–285.
https://doi.org/10.1016/j.envsoft.2013.09.031
Garré, S., Coteur, I., Wongleecharoen, C., Kongkaew, T., Diels, J., Vanderborght, J.,
2013. Noninvasive Monitoring of Soil Water Dynamics in Mixed Cropping
Systems: A Case Study in Ratchaburi Province, Thailand. Vadose Zo. J. 12, 0.
https://doi.org/10.2136/vzj2012.0129
Garré, S., Javaux, M., Vanderborght, J., Pagès, L., Vereecken, H., 2011. Three-
Dimensional Electrical Resistivity Tomography to Monitor Root Zone Water
Dynamics. Vadose Zo. J. 10, 412. https://doi.org/10.2136/vzj2010.0079
Gasperikova, E., Hubbard, S.S., Watson, D.B., Baker, G.S., Peterson, J.E., Kowalsky,
M.B., Smith, M., Brooks, S., 2012. Long-term electrical resistivity monitoring of
recharge-induced contaminant plume behavior. J. Contam. Hydrol. 142–143, 33–
49. https://doi.org/10.1016/j.jconhyd.2012.09.007
Gelhar, L., Axness, C., 1983. Three-dimensional stochastic analysis of macrodispersion
in aquifers. Water Resour. Res. 19, 161–180.
Gélis, C., Revil, A., Cushing, M.E., Jougnot, D., Lemeille, F., Cabrera, J., de Hoyos, A.,
Rocher, M., 2010. Potential of electrical resistivity tomography to detect fault
zones in limestone and argillaceous formations in the experimental platform of
Tournemire, France. Pure Appl. Geophys. 167, 1405–1418.
https://doi.org/10.1007/s00024-010-0097-x
Gephart, R.E., 2010. A short history of waste management at the Hanford Site. Phys.
Chem. Earth, Parts A/B/C 35, 298–306. https://doi.org/10.1016/J.PCE.2010.03.032
Gephart, R.E., 2003. A short history of Hanford waste generation, storage, and release.
Pnnl-13605 1–39.
Bibliography
271
Ghorbanidehno, H., Kokkinaki, A., Li, J.Y., Darve, E., Kitanidis, P.K., 2015. Real-time
data assimilation for large-scale systems: The spectral Kalman filter. Adv. Water
Resour. 86, 260–272. https://doi.org/10.1016/j.advwatres.2015.07.017
Glaser, D.R., Fink, J.B., Levitt, M.T., Rucker, D.F., 2008. A summary of recent
geophysical investigations at the Department of Energy Hanford Nuclear Facility,
in: 2008 Chapman Conference on Biogeophysics. Portland, Maine.
Glover, P.W.J., 2017. A new theoretical interpretation of Archie’s saturation exponent.
Solid Earth 8, 805–816. https://doi.org/10.5194/se-8-805-2017
Glover, P.W.J., 2015. Geophysical Properties of the Near Surface Earth: Electrical
Properties, in: Treatise on Geophysics. Elsevier, pp. 89–137.
https://doi.org/10.1016/B978-0-444-53802-4.00189-5
Gottschalk, I.P., Hermans, T., Knight, R., Caers, J., Cameron, D.A., Regnery, J., McCray,
J.E., 2017. Integrating non-colocated well and geophysical data to capture
subsurface heterogeneity at an aquifer recharge and recovery site. J. Hydrol. 555,
407–419. https://doi.org/10.1016/j.jhydrol.2017.10.028
Gouveia, W.P., Scales, J.A., 1997. Resolution of seismic waveform inversion: Bayes
versus Occam. Inverse Probl. 13, 323–349. https://doi.org/10.1088/0266-
5611/13/2/009
Green, C.T., Liao, L., Nolan, B.T., Juckem, P.F., Shope, C.L., Tesoriero, A.J., Jurgens,
B.C., 2018. Regional Variability of Nitrate Fluxes in the Unsaturated Zone and
Groundwater, Wisconsin, USA. Water Resour. Res. 54, 301–322.
https://doi.org/10.1002/2017WR022012
Gunning, J., Glinsky, M.E., Hedditch, J., 2010. Resolution and uncertainty in 1D CSEM
inversion: A Bayesian approach and open-source implementation. Geophysics 75,
F151. https://doi.org/10.1190/1.3496902
Günther, T., Rücker, C., Spitzer, K., 2006. Three-dimensional modelling and inversion
of dc resistivity data incorporating topography - II. Inversion. Geophys. J. Int. 166,
506–517. https://doi.org/10.1111/j.1365-246X.2006.03011.x
Haarder, E.B., Binley, A., Looms, M.C., Doetsch, J., Nielsen, L., Jensen, K.H., 2012.
Comparing Plume Characteristics Inferred from Cross-Borehole Geophysical
Data. Vadose Zo. J. 11, 1–40. https://doi.org/10.2136/vzj2012.0031
Hadamard, J., 1902. Sur les problèmes aux dérivées partielles et leur signification
physique. Princet. Univ. Bull. 13, 49–52.
Hadley, P.W., Newell, C.J., 2012. Groundwater Remediation: The Next 30 Years.
Ground Water 50, 669–678. https://doi.org/10.1111/j.1745-6584.2012.00942.x
Hammond, G.E., Lichtner, P.C., 2010. Field-scale model for the natural attenuation of
uranium at the Hanford 300 Area using high-performance computing. Water
Resour. Res. 46, 1–31. https://doi.org/10.1029/2009WR008819
Bibliography
272
Hammond, G.E., Lichtner, P.C., Mills, R.T., 2014. Evaluating the performance of
parallel subsurface simulators: An illustrative example with PFLOTRAN. Water
Resour. Res. 50, 208–228. https://doi.org/10.1002/2012WR013483
Hayley, K., Pidlisecky, A., Bentley, L.R., 2011. Simultaneous time-lapse electrical
resistivity inversion. J. Appl. Geophys. 75, 401–411.
https://doi.org/10.1016/j.jappgeo.2011.06.035
Hen-Jones, R.M., Hughes, P.N., Stirling, R.A., Glendinning, S., Chambers, J.E., Gunn,
D.A., Cui, Y.J., 2017. Seasonal effects on geophysical–geotechnical relationships
and their implications for electrical resistivity tomography monitoring of slopes.
Acta Geotech. 12, 1159–1173. https://doi.org/10.1007/s11440-017-0523-7
Hermans, T., Kemna, A., Nguyen, F., 2016a. Covariance-constrained difference
inversion of time-lapse electrical resistivity tomography data. GEOPHYSICS 81,
E311–E322. https://doi.org/10.1190/geo2015-0491.1
Hermans, T., Nguyen, F., Caers, J., 2015. Uncertainty in training image-based inversion
of hydraulic head data constrained to ERT data:Workflow and case study. Water
Resour. Res. 51, 5332–5352. https://doi.org/10.1002/2014WR016259
Hermans, T., Nguyen, F., Klepikova, M., Dassargues, A., Caers, J., 2018. Uncertainty
Quantification of Medium-Term Heat Storage From Short-Term Geophysical
Experiments Using Bayesian Evidential Learning. Water Resour. Res. 54, 2931–
2948. https://doi.org/10.1002/2017WR022135
Hermans, T., Nguyen, F., Robert, T., Revil, A., 2014. Geophysical methods for
monitoring temperature changes in shallow low enthalpy geothermal systems.
Energies 7, 5083–5118. https://doi.org/10.3390/en7085083
Hermans, T., Oware, E., Caers, J., 2016b. Direct prediction of spatially and temporally
varying physical properties from time-lapse electrical resistance data. Water
Resour. Res. 52, 7262–7283. https://doi.org/10.1002/2016WR019126
Hermans, T., Vandenbohede, A., Lebbe, L., Martin, R., Kemna, A., Beaujean, J., Nguyen,
F., 2012. Imaging artificial salt water infiltration using electrical resistivity
tomography constrained by geostatistical data. J. Hydrol. 438–439, 168–180.
https://doi.org/10.1016/J.JHYDROL.2012.03.021
Hinnell, A.C., Ferr, T.P.A., Vrugt, J.A., Huisman, J.A., Moysey, S., Rings, J., Kowalsky,
M.B., 2010. Improved extraction of hydrologic information from geophysical data
through coupled hydrogeophysical inversion. Water Resour. Res. 46, 1–14.
https://doi.org/10.1029/2008WR007060
Hu, L.Y., Chugunova, T., 2008. Multiple-point geostatistics for modeling subsurface
heterogeneity: A comprehensive review. Water Resour. Res. 44, 1–14.
https://doi.org/10.1029/2008WR006993
Huang, H.-T., 2013. Radionuclide Migration in Low-level Radioactive Waste Disposal
at Lan-Yu Site (in Chinese). National Cheng Kung University.
Bibliography
273
Hübner, R., Heller, K., Günther, T., Kleber, A., 2015. Monitoring hillslope moisture
dynamics with surface ERT for enhancing spatial significance of hydrometric
point measurements. Hydrol. Earth Syst. Sci. 19, 225–240.
https://doi.org/10.5194/hess-19-225-2015
Huisman, J.A., Rings, J., Vrugt, J.A., Sorg, J., Vereecken, H., 2010. Hydraulic properties
of a model dike from coupled Bayesian and multi-criteria hydrogeophysical
inversion. J. Hydrol. 380, 62–73. https://doi.org/10.1016/j.jhydrol.2009.10.023
Huntley, D., Bobrowsky, P., Hendry, M., Macciotta, R., Elwood, D., Sattler, K., Best, M.,
Chambers, J., Meldrum, P., 2019. Application of multi-dimensional electrical
resistivity tomography datasets to investigate a very slow-moving landslide near
Ashcroft, British Columbia, Canada. Landslides 16, 1033–1042.
https://doi.org/10.1007/s10346-019-01147-1
Iglesias, M., Yang, Y., n.d. A Computational efficient framework for ensemble Kalman
inversion with applications to non-destructive testing and imaging. Prep.
Iglesias, M.A., 2016. A regularizing iterative ensemble Kalman method for PDE-
constrained inverse problems. Inverse Probl. 32, 025002.
https://doi.org/10.1088/0266-5611/32/2/025002
Iglesias, M.A., Lin, K., Stuart, A.M., 2014. Well-posed Bayesian geometric inverse
problems arising in subsurface flow. Inverse Probl. 30, 114001.
https://doi.org/10.1088/0266-5611/30/11/114001
Iglesias, M.A., McLaughlin, D., 2011. Level-set techniques for facies identification in
reservoir modeling. Inverse Probl. 27, 36. https://doi.org/10.1088/0266-
5611/27/3/035008
Iglesias, M.A., Stuart, A.M., 2014. Inverse problems and uncertainty quantification.
SIAM News 2–3.
Iooss, B., 2019. Package ‘sensitivity.’
Irving, J., Singha, K., 2010. Stochastic inversion of tracer test and electrical geophysical
data to estimate hydraulic conductivities. Water Resour. Res. 46, W11514.
https://doi.org/10.1029/2009WR008340
ITRC, 2010. Use and Measurement of Mass Flux and Mass Discharge, Use and
Measurement of Mass Flux and Mass Discharge. Washington DC.
JafarGandomi, A., Binley, A., 2013. A Bayesian trans-dimensional approach for the
fusion of multiple geophysical datasets. J. Appl. Geophys. 96, 38–54.
https://doi.org/10.1016/j.jappgeo.2013.06.004
Jardani, A., Revil, A., Dupont, J.P., 2013. Stochastic joint inversion of hydrogeophysical
data for salt tracer test monitoring and hydraulic conductivity imaging. Adv.
Water Resour. 52, 62–77. https://doi.org/10.1016/j.advwatres.2012.08.005
Jayawickreme, D.H., van Dam, R.L., Hyndman, D.W., 2008. Subsurface imaging of
Bibliography
274
vegetation, climate, and root-zone moisture interactions. Geophys. Res. Lett. 35,
1–5. https://doi.org/10.1029/2008GL034690
Johnson, T., Triplett, M., Wellman, D., Northwest, P., 2014. Modeling of electrical
resistivity data in the presence of electrically conductive well casings and waste
storage tanks, in: WM2014 Conference, March 2 – 6. Phoenix, Arizona, USA, p. 7.
Johnson, T.C., Hammond, G.E., Chen, X., 2017. PFLOTRAN-E4D: A parallel open
source PFLOTRAN module for simulating time-lapse electrical resistivity data.
Comput. Geosci. 99, 72–80. https://doi.org/10.1016/j.cageo.2016.09.006
Johnson, T.C., Rucker, D.F., Glaser, D.R., 2015a. Near-Surface Geophysics at the
Hanford Nuclear Site, the United States, in: Treatise on Geophysics. Elsevier, pp.
571–595. https://doi.org/10.1016/B978-0-444-53802-4.00205-0
Johnson, T.C., Slater, L.D., Ntarlagiannis, D., Day-Lewis, F.D., Elwaseif, M., 2012a.
Monitoring groundwater-surface water interaction using time-series and time-
frequency analysis of transient three-dimensional electrical resistivity changes.
Water Resour. Res. 48, 1–13. https://doi.org/10.1029/2012WR011893
Johnson, T.C., Versteeg, R., Thomle, J., Hammond, G., Chen, X., Zachara, J., 2015b.
Four-dimensional electrical conductivity monitoring of stage-driven river water
intrusion: Accounting for water table effects using a transient mesh boundary and
conditional inversion constraints. Water Resour. Res. 51, 6177–6196.
https://doi.org/10.1002/2014WR016129
Johnson, T.C., Versteeg, R.J., Day-Lewis, F.D., Major, W., Lane, J.W., 2015c. Time-Lapse
Electrical Geophysical Monitoring of Amendment-Based Biostimulation.
Groundwater 53, 920–932. https://doi.org/10.1111/gwat.12291
Johnson, T.C., Versteeg, R.J., Huang, H., Routh, P.S., 2009. Data-domain correlation
approach for joint hydrogeologic inversion of time-lapse hydrogeologic and
geophysical data. Geophysics 74, F127. https://doi.org/10.1190/1.3237087
Johnson, T.C., Versteeg, R.J., Rockhold, M., Slater, L.D., Ntarlagiannis, D., Greenwood,
W.J., Zachara, J., 2012b. Characterization of a contaminated wellfield using 3D
electrical resistivity tomography implemented with geostatistical, discontinuous
boundary, and known conductivity constraints. Geophysics 77, EN85.
https://doi.org/10.1190/geo2012-0121.1
Johnson, T.C., Versteeg, R.J., Ward, A., Day-Lewis, F.D., Revil, A., 2010. Improved
hydrogeophysical characterization and monitoring through parallel modeling
and inversion of time-domain resistivity and induced-polarization data.
Geophysics 75, WA27. https://doi.org/10.1190/1.3475513
Johnson, T.C., Wellman, D., 2015. Accurate modelling and inversion of electrical
resistivity data in the presence of metallic infrastructure with known location and
dimension. Geophys. J. Int. 202, 1096–1108. https://doi.org/10.1093/gji/ggv206
Johnson, T.C., Wellman, D.M., 2013. Re-inversion of surface electrical resistivity
Bibliography
275
tomography data from the Hanford Site B-Complex.
Jonard, F., Bogena, H., Caterina, D., Garré, S., Klotzsche, A., Monerris, A., Schwank,
M., von Hebel, C., 2018. Ground-Based Soil Moisture Determination, in: Li, X.,
Vereecken, H. (Eds.), Observation and Measurement of Ecohydrological
Processes. Springer, pp. 1–42. https://doi.org/10.1007/978-3-662-47871-4_2-1
Jordi, C., Doetsch, J., Günther, T., Schmelzbach, C., Robertsson, J.O., 2018.
Geostatistical regularization operators for geophysical inverse problems on
irregular meshes. Geophys. J. Int. 213, 1374–1386.
https://doi.org/10.1093/gji/ggy055
Ju, L., Zhang, J., Meng, L., Wu, L., Zeng, L., 2018. An adaptive Gaussian process-based
iterative ensemble smoother for data assimilation. Adv. Water Resour.
https://doi.org/10.1016/j.advwatres.2018.03.010
Kaipio, J.P., Karjalainen, P.A., Somersalo, E., Vauhkonen, M., 1999. State estimation in
time-varying electrical impedance tomography. Ann. N. Y. Acad. Sci.
https://doi.org/10.1111/j.1749-6632.1999.tb09492.x
Kaipio, J.P., Somersalo, E., 2006. Statistical and computational inverse problems.
Springer-Verlag, New York.
Kang, X., Shi, X., Deng, Y., Revil, A., Xu, H., Wu, J., 2018. Coupled hydrogeophysical
inversion of DNAPL source zone architecture and permeability field in a 3D
heterogeneous sandbox by assimilation time-lapse cross-borehole electrical
resistivity data via ensemble Kalman filtering. J. Hydrol. 567, 149–164.
https://doi.org/10.1016/J.JHYDROL.2018.10.019
Kang, X., Shi, X., Revil, A., Cao, Z., Li, L., Lan, T., Wu, J., 2019. Coupled
hydrogeophysical inversion to identify non-Gaussian hydraulic conductivity
field by jointly assimilating geochemical and time-lapse geophysical data. J.
Hydrol. https://doi.org/10.1016/j.jhydrol.2019.124092
Karaoulis, M., Kim, J.-H., Tsourlos, P.I., 2011a. 4D active time constrained resistivity
inversion. J. Appl. Geophys. 73, 25–34.
https://doi.org/10.1016/j.jappgeo.2010.11.002
Karaoulis, M., Revil, A., Werkema, D.D., Minsley, B.J., Woodruff, W.F., Kemna, A.,
2011b. Time-lapse three-dimensional inversion of complex conductivity data
using an active time constrained (ATC) approach. Geophys. J. Int. 187, 237–251.
https://doi.org/10.1111/j.1365-246X.2011.05156.x
Kemna, A., 2000. Tomographic inversion of complex resistivity. Ruhr-Universität
Bochum.
Kemna, A., Binley, A.M., Slater, L., 2004. Cross-borehole IP imaging for engineering
and environmental applications. Geophysics 69, 97–107.
https://doi.org/10.1190/1.1649379
Key, K.T., 1977. Nuclear waster tank and pipeline external leak detection systems.
Bibliography
276
Richland, Washington.
Killick, R., Fearnhead, P., Eckley, I.A., 2012. Optimal Detection of Changepoints With
a Linear Computational Cost. J. Am. Stat. Assoc. 107, 1590–1598.
https://doi.org/10.1080/01621459.2012.737745
Kim, J.-H., Yi, M.-J., Park, S.-G., Kim, J.G., 2009. 4-D inversion of DC resistivity
monitoring data acquired over a dynamically changing earth model. J. Appl.
Geophys. 68, 522–532. https://doi.org/10.1016/j.jappgeo.2009.03.002
Kim, J., Supper, R., Ottowitz, D., Jochum, B., Yi, M., 2016. A new measurement protocol
of direct current resistivity data. Geophysics 81.
Kimbell, G.S., 1994. The interpretation of gravity and aeromagnetic data for the
geological investigation of the Sellafield area, west Cumbria. Proc. Yorksh. Geol.
Soc. 50, 103–112. https://doi.org/10.1144/pygs.50.1.103
Kitanidis, P.K., 2015a. Three-dimensional ERT imaging by the geostatistical approach,
in: American Geophysical Union Fall Meeting. San Francisco, California.
Kitanidis, P.K., 2015b. Persistent questions of heterogeneity, uncertainty, and scale in
subsurface flow and transport. Water Resour. Res. 51, 5888–5904.
https://doi.org/10.1002/2015WR017639
Kitanidis, P.K., 2011. Bayesian and Geostatistical Approaches to Inverse Problems, in:
Biegler, L., Biros, G., Ghattas, O., Heinkenschloss, M., Keyes, D., Mallick, B.,
Marzouk, Y., Tenorio, L., van Bloemen Waanders, B., Willcox, K. (Eds.), Large-
Scale Inverse Problems and Quantification of Uncertainty. John Wiley & Sons, Ltd,
Chichester, UK, pp. 71–85. https://doi.org/10.1002/9780470685853
Kitanidis, P.K., 1995. Quasi-linear geostatistical theory for inversing. Water Resour.
Res. 31, 2411–2419.
Koch, J., Nowak, W., 2016. Identification of contaminant source architectures—A
statistical inversion that emulates multiphase physics in a computationally
practicable manner. Water Resour. Res. 52. https://doi.org/10.1002/2014WR015716
Koestel, J., Kemna, A., Javaux, M., Binley, A., Vereecken, H., 2008. Quantitative
imaging of solute transport in an unsaturated and undisturbed soil monolith with
3-D ERT and TDR. Water Resour. Res. 44, 1–17.
https://doi.org/10.1029/2007WR006755
Köpke, C., Elsheikh, A.H., Irving, J., 2019. Hydrogeophysical Parameter Estimation
Using Iterative Ensemble Smoothing and Approximate Forward Solvers. Front.
Environ. Sci. 7, 34. https://doi.org/10.3389/fenvs.2019.00034
Kowalsky, M.B., Finsterle, S., Peterson, J., Hubbard, S., Rubin, Y., Majer, E., Ward, A.,
Gee, G., 2005. Estimation of field-scale soil hydraulic and dielectric parameters
through joint inversion of GPR and hydrological data. Water Resour. Res. 41, n/a-
n/a. https://doi.org/10.1029/2005WR004237
Bibliography
277
Kowalsky, M.B., Gasperikova, E., Finsterle, S., Watson, D., Baker, G., Hubbard, S.S.,
2011. Coupled modeling of hydrogeochemical and electrical resistivity data for
exploring the impact of recharge on subsurface contamination. Water Resour. Res.
47, W02509. https://doi.org/10.1029/2009WR008947
Kuras, O., Pritchard, J.D., Meldrum, P.I., Chambers, J.E., Wilkinson, P.B., Ogilvy, R.D.,
2009. Monitoring hydraulic processes with Automated time-Lapse Electrical
Resistivity Tomography ( ALERT ). Comptes Rendus Geosci. 351, 868–885.
Kuras, O., Wilkinson, P.B., Meldrum, P.I., Oxby, L.S., Uhlemann, S., Chambers, J.E.,
Binley, A., Graham, J., Smith, N.T., Atherton, N., 2016. Geoelectrical monitoring
of simulated subsurface leakage to support high-hazard nuclear
decommissioning at the Sellafield Site, UK. Sci. Total Environ. 566–567, 350–359.
https://doi.org/10.1016/j.scitotenv.2016.04.212
Kuras, O., Wilkinson, P.B., Meldrum, P.I., Oxby, L.S., Uhlemann, S., Chambers, J.E.,
Binley, A.M., Graham, J., Dewey, G., Atherton, N., 2014. A trial of 4D Cross-hole
electrical resistivity tomography (ERT) for detecting and monitoring subsurface
leakage and contaminant transport, supporting the decommissioning of legacy
silos at the Sellafield Site, UK, in: WM2014 Conference, March 2 – 6. Phoenix,
Arizona, USA.
Kuras, O., Wilkinson, P.B., Meldrum, P.I., Uhlemann, S., Chambers, J.E., 2015. Results
of the GEMS ERT trial and extended baseline measurements at Sellafield.
Keyworth.
Kuras, O., Wilkinson, P.B., White, J.C., Chambers, J.E., Meldrum, P.I., Ogilvy, R.D.,
2011. MSSS leak mitigation-leak detection phase 3: Desk study for ERT technology.
Keyworth.
Kwong, S., Fowler, L., 2014. Technical Memorandum: MSSS Electrical Resistivity
Tomography (ERT): TRAFFIC modelling of the ERT trial.
LaBrecque, D.J., Heath, G., Sharpe, R., Versteeg, R., 2004. Autonomous Monitoring of
Fluid Movement Using 3-D Electrical Resistivity Tomography. J. Environ. Eng.
Geophys. 9, 167–176. https://doi.org/10.4133/JEEG9.3.167
LaBrecque, D.J., Mletto, M., Daily, W., Ramirez, A.L., Owen, E., 1996a. The effects of
noise on Occam’s inversion of resistivity tomography data. Geophysics 61, 538.
https://doi.org/10.1190/1.1443980
LaBrecque, D.J., Ramirez, A.L., Daily, W.D., Binley, A.M., Schima, S.A., 1996b. ERT
monitoring on environmental remediation processes. Meas. Sci. Technol. 7, 375–
383. https://doi.org/10.1088/0957-0233/7/3/019
Labrecque, D.J., Yang, X., 2001. Difference inversion of ERT data: a fast inversion
method for 3-D in situ monitoring. J. Environ. Eng. Geophys. 6, 83–89.
https://doi.org/dx.doi.org/10.4133/JEEG6.2.83
Laloy, E., Hérault, R., Jacques, D., Linde, N., 2017. Efficient training-image based
Bibliography
278
geostatistical simulation and inversion using a spatial generative adversarial
neural network 1–44.
Laloy, E., Javaux, M., Vanclooster, M., Roisin, C., Bielders, C.L., 2011. Electrical
Resistivity in a Loamy Soil: Identification of the Appropriate Pedo-Electrical
Model. Vadose Zo. J. 10, 1023. https://doi.org/10.2136/vzj2010.0095
Lan, T., Shi, X., Jiang, B., Sun, Y., Wu, J., 2018. Joint inversion of physical and
geochemical parameters in groundwater models by sequential ensemble-based
optimal design. Stoch. Environ. Res. Risk Assess. 5.
https://doi.org/10.1007/s00477-018-1521-5
Last, G. V, Horton, D.G., 2000. Review of geophysical characterization methods used
at the Hanford Site.
Le, D.H., Emerick, A.A., Reynolds, A.C., 2016. An Adaptive Ensemble Smoother With
Multiple Data Assimilation for Assisted History Matching. SPE J. 21, 2195–2207.
https://doi.org/10.2118/173214-PA
Lean, C., 1998a. The use of surface geophysics in the geological characterisation of
Sellafield. Risley, United Kingdom.
Lean, C., 1998b. TerraDat geophysical site investigation at the buried channel area of
Sellafield. Risley, United Kingdom.
Lee, J., Kitanidis, P.K., 2014. Large-scale hydraulic tomography and joint inversion of
head and tracer data using the Principal Component Geostatistical Approach
(PCGA). Water Resour. Res. 50. https://doi.org/10.1002/2014WR015483
Lee, J., Liu, X., Kitanidis, P.K., Kim, U., Parker, J., Bloom, A., Lyon, R., 2012. Cost
optimization of DNAPL remediation at Dover Air Force Base Site. Groundw.
Monit. Remediat. 48–56. https://doi.org/10.1111/j1745
Lehikoinen, A., Finsterle, S., Voutilainen, A., Kowalsky, M.B., Kaipio, J.P., 2009.
Dynamical inversion of geophysical ERT data: state estimation in the vadose zone.
Inverse Probl. Sci. Eng. 17, 715–736. https://doi.org/10.1080/17415970802475951
Lehmann, P., Gambazzi, F., Suski, B., Baron, L., Askarinejad, A., Springman, S.M.,
Holliger, K., Or, D., 2013. Evolution of soil wetting patterns preceding a
hydrologically induced landslide inferred from electrical resistivity survey and
point measurements of volumetric water content and pore water pressure. Water
Resour. Res. 49, 7992–8004. https://doi.org/10.1002/2013WR014560
Lelièvre, P.G., Farquharson, C.G., 2013. Gradient and smoothness regularization
operators for geophysical inversion on unstructured meshes. Geophys. J. Int. 195,
330–341. https://doi.org/10.1093/gji/ggt255
Lesmes, D.P., Friedman, S.P., 2005. Relationships between the Electrical and
Hydrogeological Properties of Rocks and Soils, in: Rubin, Y., Hubbard, S.S. (Eds.),
Hydrogeophysics. Springer Netherlands, Dordrecht, pp. 87–128.
https://doi.org/10.1007/1-4020-3102-5_4
Bibliography
279
Lesparre, N., Nguyen, F., Kemna, A., Robert, T., Hermans, T., Daoudi, M., Flores-
Orozco, A., 2017. A new approach for time-lapse data weighting in electrical
resistivity tomography. GEOPHYSICS 82, E325–E333.
https://doi.org/10.1190/geo2017-0024.1
Leube, P.C., Geiges, A., Nowak, W., 2012. Bayesian assessment of the expected data
impact on prediction confidence in optimal sampling design. Water Resour. Res.
48, 1–16. https://doi.org/10.1029/2010WR010137
Li, J.Y., Kokkinaki, A., Ghorbanidehno, H., Darve, E.F., Kitanidis, P.K., 2015. The
compressed state Kalman filter for nonlinear state estimation: Application to
large-scale reservoir monitoring. Water Resour. Res. 51, 9942–9963.
https://doi.org/10.1002/2015WR017203
Li, L., Puzel, R., Davis, A., 2018. Data assimilation in groundwater modelling: ensemble
Kalman filter versus ensemble smoothers. Hydrol. Process. 32, 2020–2029.
https://doi.org/10.1002/hyp.13127
Li, Y.J., Kokkinaki, A., Darve, E.F., 2016. Smoothing-based compressed state Kalman
filter for joint state-parameter estimation : applications in reservoir
characterization and CO 2 storage monitoring. Water Resour. Res. 53, 7190–7207.
https://doi.org/10.1002/2016WR020168
Linde, N., 2014. Falsification and corroboration of conceptual hydrological models
using geophysical data. Wiley Interdiscip. Rev. Water 1, 151–171.
https://doi.org/10.1002/wat2.1011
Linde, N., Binley, A., Tryggvason, A., Pedersen, L.B., Revil, A., 2006. Improved
hydrogeophysical characterization using joint inversion of cross-hole electrical
resistance and ground-penetrating radar traveltime data. Water Resour. Res. 42,
WR005131. https://doi.org/10.1029/2006WR005131
Linde, N., Doetsch, J., 2016. Joint Inversion in Hydrogeophysics and Near-Surface
Geophysics, in: Moorkamp, M., Lelievre, P., Linde, N., Khan, A. (Eds.), Integrated
Imaging of the Earth. John Wiley & Sons, Inc., Hoboken, New Jersey, pp. 117–135.
https://doi.org/10.1002/9781118929063.ch7
Linde, N., Ginsbourger, D., Irving, J., Nobile, F., Doucet, A., 2017. On uncertainty
quantification in hydrogeology and hydrogeophysics. Adv. Water Resour. 110,
166–181. https://doi.org/10.1016/J.ADVWATRES.2017.10.014
Linde, N., Lochbühler, T., Dogan, M., Van Dam, R.L., 2015a. Tomogram-based
comparison of geostatistical models: Application to the Macrodispersion
Experiment (MADE) site. J. Hydrol. 531, 543–556.
https://doi.org/10.1016/j.jhydrol.2015.10.073
Linde, N., Renard, P., Mukerji, T., Caers, J., 2015b. Geological Realism in
Hydrogeological and Geophysical Inverse Modeling: a Review. Adv. Water
Resour. 86, 86–101. https://doi.org/10.1016/j.advwatres.2015.09.019
Bibliography
280
Liu, X., Lee, J., Kitanidis, P.K., Parker, J., Kim, U., 2012. Value of information as a
context-specific measure of uncertainty in groundwater remediation. Water
Resour. Manag. 26, 1513–1535. https://doi.org/10.1007/s11269-011-9970-3
Lochbühler, T., Doetsch, J., Brauchler, R., Linde, N., 2013. Structure-coupled joint
inversion of geophysical and hydrological data. Geophysics 78, ID1–ID14.
https://doi.org/10.1190/geo2012-0460.1
Loke, M.H., Barker, R.D., 1996. Practical techniques for 3D resistivity surveys and data
inversion. Geophys. Prospect. 44, 499–523. https://doi.org/10.1111/j.1365-
2478.1996.tb00162.x
Loke, M.H., Wilkinson, P.B., Uhlemann, S.S., Chambers, J.E., Oxby, L.S., 2014.
Computation of optimized arrays for 3-D electrical imaging surveys. Geophys. J.
Int. 199, 1751–1764. https://doi.org/10.1093/gji/ggu357
Looms, M.C., Binley, A., Jensen, K.H., Nielsen, L., Hansen, T.M., 2008. Identifying
Unsaturated Hydraulic Parameters Using an Integrated Data Fusion Approach
on Cross-Borehole Geophysical Data. Vadose Zo. J. 7, 238.
https://doi.org/10.2136/vzj2007.0087
Lytle, R.J., Dines, K.A., 1978. An impedance camera : A system for determining the
spatial variation of electrical conductivity. Lawrence Livermore Natl. Lab. Distrib.
16.
Manoli, G., Rossi, M., Pasetto, D., Deiana, R., Ferraris, S., Cassiani, G., Putti, M., 2015.
An iterative particle filter approach for coupled hydro-geophysical inversion of a
controlled infiltration experiment. J. Comput. Phys. 283, 37–51.
https://doi.org/10.1016/J.JCP.2014.11.035
Mariethoz, G., Caers, J., 2014. Multiple-Point Geostatistics. John Wiley & Sons, Ltd,
Chichester, UK.
Mariethoz, G., Renard, P., Straubhaar, J., 2010. The direct sampling method to perform
multiple-point geostatistical simulations. Water Resour. Res. 46, 1–14.
https://doi.org/10.1029/2008WR007621
Martini, E., Werban, U., Zacharias, S., Pohle, M., Dietrich, P., Wollschläger, U., 2017.
Repeated electromagnetic induction measurements for mapping soil moisture at
the field scale: validation with data from a wireless soil moisture monitoring
network. Hydrol. Earth Syst. Sci. 21, 495–513. https://doi.org/10.5194/hess-21-495-
2017
Maurer, H., Boerner, D.E., Curtis, A., 2000. Design strategies for electromagnetic
geophysical surveys. Inverse Probl. 16, 1097–1117. https://doi.org/10.1088/0266-
5611/16/5/302
Maurer, H., Curtis, A., Boerner, D.E., 2010. Recent advances in optimized geophysical
survey design. Geophysics 75, A177--A194. https://doi.org/10.1190/1.3484194
Maurya, P.K., Rønde, V.K., Fiandaca, G., Balbarini, N., Auken, E., Bjerg, P.L.,
Bibliography
281
Christiansen, A.V., 2017. Detailed landfill leachate plume mapping using 2D and
3D electrical resistivity tomography - with correlation to ionic strength measured
in screens. J. Appl. Geophys. 138, 1–8.
https://doi.org/10.1016/j.jappgeo.2017.01.019
McLachlan, P.J., Chambers, J.E., Uhlemann, S.S., Binley, A., 2017. Geophysical
characterisation of the groundwater–surface water interface. Adv. Water Resour.
109, 302–319. https://doi.org/10.1016/j.advwatres.2017.09.016
Menke, W., 1989. Geophysical Data Analysis: Discrete Inverse Theory. Academic Press.
Michalak, A.M., Shlomi, S., 2007. A geostatistical data assimilation approach for
estimating groundwater plume distributions from multiple monitoring events.
pp. 73–88. https://doi.org/10.1029/171GM08
Michot, D., Benderitter, Y., Dorigny, A., Nicoullaud, B., King, D., Tabbagh, A., 2003.
Spatial and temporal monitoring of soil water content with an irrigated corn crop
cover using surface electrical resistivity tomography. Water Resour. Res. 39, 1–20.
https://doi.org/10.1029/2002WR001581
Millard, S.P., 2013. Designing a Sampling Program. EnvStats An R Packag. Environ.
Stat. 25–61. https://doi.org/10.1007/978-1-4614-8456-1
Miller, C.R., Routh, P.S., 2007. Resolution analysis of geophysical images: Comparison
between point spread function and region of data influence measures. Geophys.
Prospect. 55, 835–852. https://doi.org/10.1111/j.1365-2478.2007.00640.x
Miller, C.R., Routh, P.S., Brosten, T.R., McNamara, J.P., 2008. Application of time-lapse
ERT imaging to watershed characterization. Geophysics 73, G7.
https://doi.org/10.1190/1.2907156
Minsley, B.J., Sogade, J., Morgan, F.D., 2007. Three-dimensional self-potential
inversion for subsurface DNAPL contaminant detection at the Savannah River
Site, South Carolina. Water Resour. Res. 43, 1–13.
https://doi.org/10.1029/2005WR003996
Mitchell, M.A., Oldenburg, D.W., 2016. Data quality control methodology for large ,
non-conventional DC resistivity datasets. J. Appl. Geophys. 3, 948–953.
https://doi.org/10.1016/j.jappgeo.2016.09.018
Mitchell, V., Knight, R., Pidlisecky, A., 2011. Inversion of time-lapse electrical
resistivity imaging data for monitoring infiltration. Lead. Edge 140–144.
Moghadas, D., Badorreck, A., 2019. Machine learning to estimate soil moisture from
geophysical measurements of electrical conductivity. Near Surf. Geophys.
https://doi.org/10.1002/nsg.12036
Morelli, G., LaBrecque, D.J., 1996. Advances in ERT inverse modeling. Eur. J. Environ.
Eng. Geophys. 1, 171–186.
Morris, M.D., 1991. Factorial Sampling Plans for Preliminary Computational
Bibliography
282
Experiments, Technometrics.
Morrison, H.F., Beckers, A., Lee, K.H., 1987. Measurements of electrical conductivity
for characterizing and monitoring nuclear waste repositories. Berkeley, California,
United States.
Mosegaard, K., 1999. Resolution analysis of general inverse problems through inverse
Monte Carlo sampling. Inverse Probl. 14, 405–426. https://doi.org/10.1088/0266-
5611/14/3/004
Mosegaard, K., Tarantola, A., 1995. Monte Carlo sampling of solutions to inverse
problems. J. Geophys. Res. (Solid Earth) 100, 12431–12448.
https://doi.org/10.1029/94JB03097
Moysey, S., Singha, K., Knight, R., 2005. A framework for inferring field-scale rock
physics relationships through numerical simulation. Geophys. Res. Lett. 32, 1–4.
https://doi.org/10.1029/2004GL022152
Murakami, H., Chen, X., Hahn, M.S., Liu, Y., Rockhold, M.L., Vermeul, V.R., Zachara,
J.M., Rubin, Y., 2010. Bayesian approach for three-dimensional aquifer
characterization at the Hanford 300 Area. Hydrol. Earth Syst. Sci. 14, 1989–2001.
https://doi.org/10.5194/hess-14-1989-2010
Musgrave, H., Binley, A., 2011. Revealing the temporal dynamics of subsurface
temperature in a wetland using time-lapse geophysics. J. Hydrol. 396, 258–266.
https://doi.org/10.1016/j.jhydrol.2010.11.008
Mwakanyamale, K., Slater, L., Binley, A., Ntarlagiannis, D., 2012. Lithologic imaging
using complex conductivity: Lessons learned from the Hanford 300 Area.
Geophysics 77, E397. https://doi.org/10.1190/geo2011-0407.1
Myers, D.A., Cubbage, B., Brauchla, R., O’Brien, G., Bergeron, M., Henderson, C., 2008a.
Surface geophyscial exploration of TX-TY tank farms at the Hanford Site: Results
of background characterization with ground penetrating radar. RPP-RPT-38104.
Richland, Washington.
Myers, D.A., Rucker, D.F., Levitt, M., Cubbage, B., Henderson, C., 2008b. Surface
geophyscial exploration of TX-TY tank farms at the Hanford Site: Results of
background characterization with magnetics and electromagnetics. RPP-RPT-
36893. Richland, Washington.
Nearing, G.S., Gupta, H. V., 2015. The quantity and quality of information in
hydrologic models. Water Resour. Res. 51, 524–538.
https://doi.org/10.1002/2014WR015895
Nenna, V., Knight, R., 2013. Demonstration of a value of information metric to assess
the use of geophysical data for a groundwater application. Geophysics 79, E51–
E60. https://doi.org/10.1190/geo2012-0474.1
Nenna, V., Pidlisecky, A., Knight, R., 2011. Informed experimental design for electrical
resistivity imaging. Near Surf. Geophys. 9, 469–482. https://doi.org/10.3997/1873-
Bibliography
283
0604.2011027
Neuman, S.P., Xue, L., Ye, M., Lu, D., 2012. Bayesian analysis of data-worth
considering model and parameter uncertainties. Adv. Water Resour. 36, 75–85.
https://doi.org/10.1016/j.advwatres.2011.02.007
Nguyen, F., Garambois, S., Jongmans, D., Pirard, E., Loke, M.H., 2005. Image
processing of 2D resistivity data for imaging faults. J. Appl. Geophys. 57, 260–277.
https://doi.org/10.1016/j.jappgeo.2005.02.001
Nguyen, F., Kemna, A., Robert, T., Hermans, T., 2016. Data-driven selection of the
minimum-gradient support parameter in time-lapse focused electric imaging.
GEOPHYSICS 81, A1–A5. https://doi.org/10.1190/geo2015-0226.1
Nimmer, R.E., Osiensky, J.L., Binley, A.M., Williams, B.C., 2008. Three-dimensional
effects causing artifacts in two-dimensional, cross-borehole, electrical imaging. J.
Hydrol. 359, 59–70. https://doi.org/10.1016/j.jhydrol.2008.06.022
Nolet, G., Montelli, R., Virieux, J., 1999. Explicit, approximate expressions for the
resolution and a posteriori covariance of massive tomographic systems. Geophys.
J. Int. 138, 36–44. https://doi.org/10.1046/j.1365-246X.1999.00858.x
Norton, M.G., Arthur, J.C.R., Dyer, K.J., 1997. Geophysical survey planning for the
Dounreay and Sellafield geological investigations. Geol. Soc. London, Eng. Geol.
Spec. Publ. 12, 335–343. https://doi.org/10.1144/GSL.ENG.1997.012.01.31
Nowak, W., 2009. Best unbiased ensemble linearization and the quasi-linear Kalman
ensemble generator. Water Resour. Res. 45.
https://doi.org/10.1029/2008WR007328
Nowak, W., Bode, F., Loschko, M., 2015. A Multi-objective Optimization Concept for
Risk-based Early-warning Monitoring Networks in Well Catchments. Procedia
Environ. Sci. 25, 191–198. https://doi.org/10.1016/j.proenv.2015.04.026
Nowak, W., de Barros, F.P.J., Rubin, Y., 2010. Bayesian geostatistical design: Task-
driven optimal site investigation when the geostatistical model is uncertain.
Water Resour. Res. 46, 1–17. https://doi.org/10.1029/2009WR008312
Nowak, W., Rubin, Y., de Barros, F.P.J., 2012. A hypothesis-driven approach to
optimize field campaigns. Water Resour. Res. 48, W06509.
https://doi.org/10.1029/2011WR011016
Nussbaumer, R., Linde, N., Mariethoz, G., Holliger, K., 2019. Simulation of fine-scale
electrical conductivity fields using resolution-limited tomograms and area-to-
point kriging. Geophys. J. Int. 218, 1322–1335. https://doi.org/10.1093/gji/ggz185
Oldenborger, G.A., Knoll, M.D., Routh, P.S., LaBrecque, D.J., 2007. Time-lapse ERT
monitoring of an injection/withdrawal experiment in a shallow unconfined
aquifer. Geophysics 72, F177–F187. https://doi.org/10.1190/1.2734365
Oldenborger, G.A., LeBlanc, A.M., 2015. Geophysical characterization of permafrost
Bibliography
284
terrain at Iqaluit International Airport, Nunavut. J. Appl. Geophys. 123, 36–49.
https://doi.org/10.1016/j.jappgeo.2015.09.016
Oldenborger, G.A., Routh, P.S., 2009. The point-spread function measure of resolution
for the 3-D electrical resistivity experiment. Geophys. J. Int. 176, 405–414.
https://doi.org/10.1111/j.1365-246X.2008.04003.x
Oldenborger, G.A., Routh, P.S., Knoll, M.D., 2005. Sensitivity of electrical resistivity
tomography data to electrode position errors. Geophys. J. Int. 163, 1–9.
https://doi.org/10.1111/j.1365-246X.2005.02714.x
Oldenburg, D.W., Li, Y., 2005. Inversion for Applied Geophysics: A Tutorial, in: Near-
Surface Geophysics. Society of Exploration Geophysicists, pp. 89–150.
https://doi.org/10.1190/1.9781560801719.ch5
Oldenburg, D.W., Li, Y., 1999. Estimating depth of investigation in dc resistivity and
IP surveys. Geophysics 64, 403. https://doi.org/10.1190/1.1444545
Olsson, O., Anderson, P., Gustafsson, E., 1991. PROJECT 9118 Site Characterization
and Validation - Monitoring of Saline Tracer Transport by Borehole Radar
Measurements, Final Report.pdf. Stockholm.
Olsson, O., Gale, J.E., 1995. Site assessment and characterization for high-level nuclear
waste disposal; results from the Stripa Project, Sweden. Q. J. Eng. Geol. 28, 17–30.
Oware, E.K., Moysey, S.M.J., Khan, T., 2013. Physically based regularization of
hydrogeophysical inverse problems for improved imaging of process-driven
systems. Water Resour. Res. 49, 6238–6247. https://doi.org/10.1002/wrcr.20462
Pacific Northwest National Laboratory, 2016. An Award-Winning Subsurface
Software [WWW Document]. URL
https://energyenvironment.pnnl.gov/highlights/highlight.asp?id=2556 (accessed
9.16.19).
Parasnis, D.S., 1988. Reciprocity theorems in geoelectric and geoelectromagnetic work.
Geoexploration 25, 177–198. https://doi.org/10.1016/0016-7142(88)90014-2
Park, H., Scheidt, C., Fenwick, D., Boucher, A., Caers, J., 2013. History matching and
uncertainty quantification of facies models with multiple geological
interpretations. Comput. Geosci. 17, 609–621. https://doi.org/10.1007/s10596-013-
9343-5
Parsekian, A.D., Claes, N., Singha, K., Minsley, B.J., Carr, B., Voytek, E., Harmon, R.,
Kass, A., Carey, A., Thayer, D., Flinchum, B., 2017. Comparing Measurement
Response and Inverted Results of Electrical Resistivity Tomography Instruments.
J. Environ. Eng. Geophys. 22, 249–266. https://doi.org/10.2113/JEEG22.3.249
Parsekian, A.D., Singha, K., Minsley, B.J., Holbrook, W.S., Slater, L., 2015. Multiscale
geophysical imaging of the critical zone. Rev. Geophys. 53, 1–26.
https://doi.org/10.1002/2014RG000465
Bibliography
285
Paté-Cornell, E., Cochran, J.J., Cox, L.A., Keskinocak, P., Kharoufeh, J.P., Smith, J.C.,
Assessment, F.R., 2010. An introduction to probabilistic risk analysis for
engineered systems. Wiley Encycl. Oper. Res. Manag. Sci.
https://doi.org/10.1002/9780470400531.eorms0680
Pearson, E.S., Hartley, H.O. (Eds.), 1970. Biometrika tables for statisticians. Cambridge
University Press.
Pidlisecky, A., Singha, K., Day-Lewis, F.D., 2011. A distribution-based parametrization
for improved tomographic imaging of solute plumes. Geophys. J. Int. 187, 214–
224. https://doi.org/10.1111/j.1365-246X.2011.05131.x
Pinheiro, J., Bates, D., 1988. Unconstrained parameterizations for variance-covariance
matrices. Stat. Comput. 6, 289–296.
Pollock, D., Cirpka, O.A., 2010. Fully coupled hydrogeophysical inversion of synthetic
salt tracer experiments. Water Resour. Res. 46, W07501.
https://doi.org/10.1029/2009WR008575
Portniaguine, O., Zhdanov, M.S., 1999. Focusing geophysical inversion images.
GEOPHYSICS 64, 874–887. https://doi.org/10.1190/1.1444596
Power, C., Gerhard, J.I., Karaoulis, M., Tsourlos, P., Giannopoulos, A., 2014. Evaluating
four-dimensional time-lapse electrical resistivity tomography for monitoring
DNAPL source zone remediation. J. Contam. Hydrol. 162–163, 27–46.
https://doi.org/10.1016/j.jconhyd.2014.04.004
Ramirez, A., Daily, W., 1996. Detection of leaks in underground storage tanks using
electrical resistance methods:1996 results. Livermore, California.
Ramirez, A., Daily, W., Binley, A., Labrecque, D., Roelant, D., 1996. Detection of leaks
in underground storage tanks using electrical resistance methods. J. Environ. Eng.
Geophys. 1, 189–203. https://doi.org/10.4133/JEEG1.3.189
Ramirez, A.L., Daily, W.D., Newmark, R.L., 1995. Electrical resistance tomography for
steam injection monitoring and process control. J. Environ. Eng. Geophys. 1, 39.
https://doi.org/10.4133/JEEG1.A.39
Ramirez, A.L., Nitao, J.J., Hanley, W.G., Aines, R., Glaser, R.E., Sengupta, S.K., Dyer,
K.M., Hickling, T.L., Daily, W.D., 2005. Stochastic inversion of electrical resistivity
changes using a Markov Chain Monte Carlo approach. J. Geophys. Res. 110,
B02101. https://doi.org/10.1029/2004JB003449
Ray, A., Kaplan, S., Washbourne, J., Albertin, U., 2018. Low frequency full waveform
seismic inversion within a tree based Bayesian framework. Geophys. J. Int. 212,
522–542. https://doi.org/10.1093/gji/ggx428
Ray, A., Sekar, A., Hoversten, G.M., Albertin, U., 2016. Frequency domain full
waveform elastic inversion of marine seismic data from the Alba field using a
Bayesian trans-dimensional algorithm. Geophys. J. Int. 205, 915–937.
https://doi.org/10.1093/gji/ggw061
Bibliography
286
Ray, R.L., Jacobs, J.M., 2007. Relationships among remotely sensed soil moisture,
precipitation and landslide events. Nat. Hazards 43, 211–222.
https://doi.org/10.1007/s11069-006-9095-9
Reichle, R.H., Crow, W.T., Keppenne, C.L., 2008. An adaptive ensemble Kalman filter
for soil moisture data assimilation. Water Resour. Res. 44, 1–13.
https://doi.org/10.1029/2007WR006357
Revil, A., Skold, M., Hubbard, S.S., Wu, Y., Watson, D.B., Karaoulis, M., 2013a.
Petrophysical properties of saprolites from the Oak Ridge Integrated Field
Research Challenge site, Tennessee. Geophysics 78, D21–D40.
https://doi.org/10.1190/geo2012-0176.1
Revil, A., Wu, Y., Karaoulis, M., Hubbard, S.S., Watson, D.B., Eppehimer, J.D., 2013b.
Geochemical and geophysical responses during the infiltration of fresh water into
the contaminated saprolite of the Oak Ridge Integrated Field Research Challenge
site, Tennessee. Water Resour. Res. 49, n/a-n/a. https://doi.org/10.1002/wrcr.20380
Reynolds, A.C., Zafari, M., Li, G., 2006. Iterative forms of the ensemble Kalman filter,
in: Proceedings of 10th European Conference on the Mathematics of Oil Recovery.
Amsterdam, The Netherlands.
Robinet, J., von Hebel, C., Govers, G., van der Kruk, J., Minella, J.P.G., Schlesner, A.,
Ameijeiras-Mariño, Y., Vanderborght, J., 2018. Spatial variability of soil water
content and soil electrical conductivity across scales derived from
Electromagnetic Induction and Time Domain Reflectometry. Geoderma 314, 160–
174. https://doi.org/10.1016/j.geoderma.2017.10.045
Robins, N.S., 1979. Electrical resistivity survey at the Windscale Works, Cumbria.
Oxfordshire, United Kingdom.
Robinson, D.A., Lebron, I., Kocar, B., Phan, K., Sampson, M., Crook, N., Fendorf, S.,
2009. Time-lapse geophysical imaging of soil moisture dynamics in tropical
deltaic soils: An aid to interpreting hydrological and geochemical processes.
Water Resour. Res. 45, n/a-n/a. https://doi.org/10.1029/2008WR006984
Robinson, J., 2015. Improving characterization of fractured rock using 3D cross-
borehole electrical resistivity tomography (ERT). Rutgers University.
https://doi.org/10.7282/T3DF6T3V
Robinson, J., Johnson, T., Slater, L., 2015. Challenges and opportunities for fractured
rock imaging using 3D cross borehole electrical resistivity. Geophysics 80, E49–
E61. https://doi.org/10.1190/GEO2014-0138.1
Rojas, R., Feyen, L., Dassargues, A., 2008. Conceptual model uncertainty in
groundwater modeling: Combining generalized likelihood uncertainty
estimation and Bayesian model averaging. Water Resour. Res. 44, 1–16.
https://doi.org/10.1029/2008WR006908
Romero-Ruiz, A., Linde, N., Keller, T., Or, D., 2018. A Review of Geophysical Methods
Bibliography
287
for Soil Structure Characterization. Rev. Geophys. 56, 672–697.
https://doi.org/10.1029/2018RG000611
Ross, M., 2004. Consideration of available geophysical techniques for non-intrusive
survey of B291 trenches. Sellafield, United Kingdom.
Rossi, M., Manoli, G., Pasetto, D., Deiana, R., Ferraris, S., Strobbia, C., Putti, M.,
Cassiani, G., 2015. Coupled inverse modeling of a controlled irrigation
experiment using multiple hydro-geophysical data. Adv. Water Resour. 82, 150–
165. https://doi.org/10.1016/J.ADVWATRES.2015.03.008
Roubinet, D., Irving, J., 2014. Discrete-dual-porosity model for electric current flow in
fractured rock. J. Geophys. Res. Solid Earth 119, 767–786.
https://doi.org/10.1002/2013JB010668
Routh, P.S., 2009. A practical strategy to interrogate resolution , uncertainty and value
of information in geophysical inverse problem. Geohorizons 27–37.
Rubin, Y., Chen, X., Murakami, H., Hahn, M., 2010. A Bayesian approach for inverse
modeling, data assimilation, and conditional simulation of spatial random fields.
Water Resour. Res. 46, 1–23. https://doi.org/10.1029/2009WR008799
Rubin, Y., Hubbard, S.S. (Eds.), 2005. Hydrogeophysics. Springer.
Rucker, D., Levitt, M., O’Brien, G., Henderson, C., 2007. Surface geophysical
exploration of B , BX , and BY Tank Farms at the Hanford Site : Results of
background characterization with ground penetrating radar.
Rucker, D.F., 2012. Enhanced resolution for long electrode ERT. Geophys. J. Int. 191,
101–111. https://doi.org/10.1111/j.1365-246X.2012.05643.x
Rucker, D.F., Myers, D.A., Cubbage, B., Levitt, M.T., Noonan, G.E., McNeill, M.,
Henderson, C., Lober, R.W., 2013. Surface geophysical exploration: Developing
noninvasive tools to monitor past leaks around Hanford’s tank farms. Environ.
Monit. Assess. 185, 995–1010. https://doi.org/10.1007/s10661-012-2609-x
Ruggeri, P., Gloaguen, E., Lefebvre, R., Irving, J., Holliger, K., 2014. Integration of
hydrological and geophysical data beyond the local scale: Application of Bayesian
sequential simulation to field data from the Saint-Lambert-de-Lauzon site,
Québec, Canada. J. Hydrol. 514, 271–280.
https://doi.org/10.1016/J.JHYDROL.2014.04.031
Ruggeri, P., Irving, J., Gloaguen, E., Holliger, K., 2013. Regional-scale integration of
multiresolution hydrological and geophysical data using a two-step Bayesian
sequential simulation approach. Geophys. J. Int. 194, 289–303.
https://doi.org/10.1093/gji/ggt067
Salamon, P., Feyen, L., 2010. Disentangling uncertainties in distributed hydrological
modeling using multiplicative error models and sequential data assimilation.
Water Resour. Res. 46, 1–20. https://doi.org/10.1029/2009WR009022
Bibliography
288
Sassen, D.S., Hubbard, S.S., Bea, S.A., Chen, J., Spycher, N., Denham, M.E., 2012.
Reactive facies: An approach for parameterizing field-scale reactive transport
models using geophysical methods. Water Resour. Res. 48, 1–20.
https://doi.org/10.1029/2011WR011047
Sato, K., 2011. Value of information analysis for adequate monitoring of carbon dioxide
storage in geological reservoirs under uncertainty. Int. J. Greenh. Gas Control 5,
1294–1302. https://doi.org/10.1016/j.ijggc.2011.07.010
Scaini, A., Audebert, M., Hissler, C., Fenicia, F., Gourdol, L., Pfister, L., Beven, K.J.,
2017. Velocity and celerity dynamics at plot scale inferred from artificial tracing
experiments and time-lapse ERT. J. Hydrol. 546, 28–43.
https://doi.org/10.1016/J.JHYDROL.2016.12.035
Scales, J.A., Snieder, R., 1997. To Bayes or not to Bayes? GEOPHYSICS 62, 1045–1046.
https://doi.org/10.1190/1.6241045.1
Schaap, M.G., Leij, F.J., van Genuchten, M.T., 2001. rosetta: a computer program for
estimating soil hydraulic parameters with hierarchical pedotransfer functions. J.
Hydrol. 251, 163–176. https://doi.org/10.1016/S0022-1694(01)00466-8
Scheidt, C., Li, L., Caers, J., 2018. Quantifying Uncertainty in Subsurface Systems,
Geophysical Monograph Series. John Wiley & Sons, Inc., Hoboken, NJ, USA.
https://doi.org/10.1002/9781119325888
Schmidt-Hattenberger, C., Bergmann, P., Labitzke, T., Wagner, F., Rippe, D., 2016.
Permanent crosshole electrical resistivity tomography (ERT) as an established
method for the long-term CO2 monitoring at the Ketzin pilot site. Int. J. Greenh.
Gas Control 52, 432–448. https://doi.org/10.1016/j.ijggc.2016.07.024
Schmidt, F., Wainwright, H.M., Faybishenko, B., Denham, M., Eddy-Dilek, C., 2018. In
Situ Monitoring of Groundwater Contamination Using the Kalman Filter.
Environ. Sci. Technol. 52, 7418–7425. https://doi.org/10.1021/acs.est.8b00017
Schnaidt, S., Heinson, G., 2015. Bootstrap resampling as a tool for uncertainty analysis
in 2-D magnetotelluric inversion modelling. Geophys. J. Int. 203, 92–106.
https://doi.org/10.1093/gji/ggv264
Scholer, M., Irving, J., Binley, A., Holliger, K., 2011. Estimating vadose zone hydraulic
properties using ground penetrating radar: The impact of prior information.
Water Resour. Res. 47, 1–14. https://doi.org/10.1029/2011WR010409
Scholer, M., Irving, J., Looms, M.C., Nielsen, L., Holliger, K., 2012. Bayesian Markov-
chain-Monte-Carlo inversion of time-lapse crosshole GPR data to characterize the
vadose zone at the Arrenaes site, Denmark. (Special Section: Model-data fusion
in the vadose zone.). Vadose Zo. J. 11, 153. https://doi.org/10.2136/vzj2011.0153
Schöniger, A., Nowak, W., Hendricks Franssen, H.-J., 2012. Parameter estimation by
ensemble Kalman filters with transformed data: Approach and application to
hydraulic tomography. Water Resour. Res. 48, W04502.
Bibliography
289
https://doi.org/10.1029/2011WR010462
Serco/Golder, 2010. Sellafield contaminated land and groundwater management
project – geophysical survey – micro-gravity, downhole logging, EM61, GPR and
ERI.
Serco/Golder, 2009. Sellafield contaminated land and groundwater management
project – Report for B291 Trench Area investigation.
Serco/Golder, 2008a. Sellafield contaminated land and groundwater management
project – geophysics survey – B291 Trench Area.
Serco/Golder, 2008b. Sellafield contaminated land and groundwater management
project – Geophysical data collection plan.
Shanahan, P.W., Binley, A., Whalley, W.R., Watts, C.W., 2015. The Use of
Electromagnetic Induction to Monitor Changes in Soil Moisture Profiles beneath
Different Wheat Genotypes. Soil Sci. Soc. Am. J.
https://doi.org/10.2136/sssaj2014.09.0360
Shen, C., 2018. A Transdisciplinary Review of Deep Learning Research and Its
Relevance for Water Resources Scientists. Water Resour. Res. 54, 8558–8593.
https://doi.org/10.1029/2018WR022643
Shi, L., Zeng, L., Tang, Y., Chen, C., Yang, J., 2013. Uncertainty quantification of
contaminant transport and risk assessment with conditional stochastic collocation
method. Stoch. Environ. Res. Risk Assess. 27, 1453–1464.
https://doi.org/10.1007/s00477-012-0682-x
Shi, W., 1998. Advanced modeling and inversion techniques for three-dimensional
geoelectrical surveys. Massachusetts Institute of Technology.
Shlomi, S., Michalak, A.M., 2007. A geostatistical framework for incorporating
transport information in estimating the distribution of a groundwater
contaminant plume. Water Resour. Res. 43.
https://doi.org/10.1029/2006WR005121
Si, H., 2015. TetGen, a Delaunay-Based Quality Tetrahedral Mesh Generator. ACM
Trans. Math. Softw. 41, 1–36. https://doi.org/10.1145/2629697
Sibbertt, L., Chambers, J.E., Li, B., Wilkinson, P.B., 2017. Total Generalised Variation:
An improved regulariser for Electrical Resistivity Tomography inversion., in:
Geophysical Research Abstracts. pp. EGU2017-14348.
Singha, K., Day-Lewis, F.D., Johnson, T., Slater, L.D., 2015. Advances in interpretation
of subsurface processes with time-lapse electrical imaging. Hydrol. Process. 29,
1549–1576. https://doi.org/10.1002/hyp.10280
Singha, K., Gorelick, S.M., 2006. Effects of spatially variable resolution on field-scale
estimates of tracer concentration from electrical inversions using Archie’s law.
Geophysics 71, G83. https://doi.org/10.1190/1.2194900
Bibliography
290
Singha, K., Moysey, S., 2006. Accounting for spatially variable resolution in electrical
resistivity tomography through field-scale rock-physics relations. Geophysics 71,
A25. https://doi.org/10.1190/1.2209753
Skyes, J.F., Wilson, J.L., Andrews, R.W., 1985. Sensitivity analysis for steady state
groundwater flow using adjoint operators. Water Resour. Res. 21, 359–371.
Slater, L., Binley, A., 2006. Synthetic and field-based electrical imaging of a zerovalent
iron barrier: Implications for monitoring long-term barrier performance.
Geophysics 71, B129–B137. https://doi.org/10.1190/1.2235931
Slater, L., Binley, A., 2003. Evaluation of permeable reactive barrier (PRB) integrity
using electrical imaging methods. Geophysics 68, 911–921.
https://doi.org/10.1190/1.1581043
Slater, L., Binley, A.M., Daily, W., Johnson, R., 2000. Cross-hole electrical imaging of a
controlled saline tracer injection. J. Appl. Geophys. 44, 85–102.
https://doi.org/10.1016/S0926-9851(00)00002-1
Slater, L.D., Binley, A., Brown, D., 1997. Electrical Imaging of Fractures Using Ground-
Water Salinity Change. Ground Water 35, 436–442. https://doi.org/10.1111/j.1745-
6584.1997.tb00103.x
Slater, L.D., Ntarlagiannis, D., Day-Lewis, F.D., Mwakanyamale, K., Versteeg, R.J.,
Ward, A., Strickland, C., Johnson, C.D., Lane, J.W., 2010. Use of electrical imaging
and distributed temperature sensing methods to characterize surface water-
groundwater exchange regulating uranium transport at the Hanford 300 Area,
Washington. Water Resour. Res. 46, 1–13. https://doi.org/10.1029/2010WR009110
Sobol′, I.., 2001. Global sensitivity indices for nonlinear mathematical models and their
Monte Carlo estimates. Math. Comput. Simul. 55, 271–280.
https://doi.org/10.1016/S0378-4754(00)00270-6
Song, X., Chen, X., Ye, M., Dai, Z., Hammond, G., Zachara, J.M., 2019. Delineating
Facies Spatial Distribution by Integrating Ensemble Data Assimilationand
Indicator Geostatistics With Level‐Set Transformation. Water Resour. Res.
2018WR023262. https://doi.org/10.1029/2018WR023262
Stenberg, L., 2008. Detailed ground magnetic and resistivity measurements for a
potential deep geologic repository for spent nuclear fuel in Oskarshamn, in:
Ahlbom, K., Stephens, M. (Eds.), Investigations of Potential Repository Sites for
Spent Nuclear Fulel at Forsmark and Laxemar-Simpevarp, Sweden. SKB,
Stockholm, p. 43.
Stordal, A.S., Elsheikh, A.H., 2015. Iterative ensemble smoothers in the annealed
importance sampling framework. Adv. Water Resour. 86, 231–239.
https://doi.org/10.1016/j.advwatres.2015.09.030
Stummer, P., Maurer, H.R., Green, A.G., 2004. Experimental design: Electrical
resistivity data sets that provide optimum subsurface information. Geophysics 69,
Bibliography
291
120–139.
Sun, A.Y., 2007. A robust geostatistical approach to contaminant source identification.
Water Resour. Res. https://doi.org/10.1029/2006WR005106
Sun, A.Y., Painter, S.L., Wittmeyer, G.W., 2006. A constrained robust least squares
approach for contaminant release history identification. Water Resour. Res. 42.
https://doi.org/10.1029/2005WR004312
Sun, N.-Z., Sun, A., 2015. Model Calibration and Parameter Estimation. Springer New
York, New York, NY. https://doi.org/10.1007/978-1-4939-2323-6
Tai, X., Chan, T.F., 2004. A survey on multiple level set methods with applications for
identifying piecewise constant functions. Int. J. Numer. Anal. Model 1, 25–48.
Tang, G., D’Azevedo, E.F., Zhang, F., Parker, J.C., Watson, D.B., Jardine, P.M., 2010.
Application of a hybrid MPI/OpenMP approach for parallel groundwater model
calibration using multi-core computers. Comput. Geosci. 36, 1451–1460.
https://doi.org/10.1016/j.cageo.2010.04.013
Tarantola, A., 2006. Popper, Bayes and the inverse problem. Nat. Phys. 2, 492–494.
https://doi.org/10.1038/nphys375
Tarantola, A., 2005. Inverse Problem Theory and Methods for Model Parameter
Estimation. SIAM.
Tartakovsky, D.M., 2013. Assessment and management of risk in subsurface
hydrology: A review and perspective. Adv. Water Resour. 51, 247–260.
https://doi.org/10.1016/j.advwatres.2012.04.007
Taylor, J.R., 1982. An Introduction to Error Analysis: The Study of Uncertainties in
Physical Measurements. University Science Books.
TERRADAT (UK) LTD., 2012. Geophysical survey to locate subsurface features and
services at B38, Sellafield, Cumbria. Cardiff, United Kingdom.
TERRADAT (UK) LTD., 2004. Geophysical investigations into the subsurface deposits
and structures. Cardiff, United Kingdom.
TERRADAT (UK) LTD., 1998. Geophysical survey trials carried out at Sellafield, West
Boundary. Cardiff, United Kingdom.
Trainor-Guitton, W.J., 2014. A geophysical perspective of value of information:
Examples of spatial decisions for groundwater sustainability. Environ. Syst. Decis.
34, 124–133. https://doi.org/10.1007/s10669-013-9487-9
Trainor-Guitton, W.J., Caers, J.K., Mukerji, T., 2011. A Methodology for Establishing a
Data Reliability Measure for Value of Spatial Information Problems. Math. Geosci.
43, 929–949. https://doi.org/10.1007/s11004-011-9367-0
Trainor-Guitton, W.J., Hoversten, G.M., Ramirez, A., Roberts, J., Juliusson, E., Key, K.,
Mellors, R., 2014. The value of spatial information for determining well placement:
Bibliography
292
A geothermal example. GEOPHYSICS 79, W27–W41.
https://doi.org/10.1190/geo2013-0337.1
Trainor-Guitton, W.J., Mukerji, T., Knight, R., 2013a. A methodology for quantifying
the value of spatial information for dynamic Earth problems. Stoch. Environ. Res.
Risk Assess. 27, 969–983. https://doi.org/10.1007/s00477-012-0619-4
Trainor-Guitton, W.J., Ramirez, A., Yang, X., Mansoor, K., Sun, Y., Carroll, S., 2013b.
Value of information methodology for assessing the ability of electrical resistivity
to detect CO2/brine leakage into a shallow aquifer. Int. J. Greenh. Gas Control 18,
101–113. https://doi.org/10.1016/j.ijggc.2013.06.018
Tran, A.P., Dafflon, B., Hubbard, S.S., 2017. Coupled land surface–subsurface
hydrogeophysical inverse modeling to estimate soil organic carbon content and
explore associated hydrological and thermal dynamics in the Arctic tundra.
Cryosph. 11, 2089–2109. https://doi.org/10.5194/tc-11-2089-2017
Tran, A.P., Dafflon, B., Hubbard, S.S., Kowalsky, M.B., Long, P., Tokunaga, T.K.,
Williams, K.H., 2016. Quantifying shallow subsurface water and heat dynamics
using coupled hydrological-thermal-geophysical inversion. Hydrol. Earth Syst.
Sci. 20, 3477–3491. https://doi.org/10.5194/hess-20-3477-2016
Trigo, F.C., Gonzalez-Lima, R., Amato, M.B.P., 2004. Electrical impedance tomography
using the extended Kalman filter. IEEE Trans. Biomed. Eng. 51, 72–81.
https://doi.org/10.1109/TBME.2003.820389
Troldborg, M., Nowak, W., Lange, I. V., Santos, M.C., Binning, P.J., Bjerg, P.L., 2012.
Application of Bayesian geostatistics for evaluation of mass discharge uncertainty
at contaminated sites. Water Resour. Res. 48, 1–19.
https://doi.org/10.1029/2011WR011785
Troldborg, M., Nowak, W., Tuxen, N., Bjerg, P.L., Helmig, R., Binning, P.J., 2010.
Uncertainty evaluation of mass discharge estimates from a contaminated site
using a fully Bayesian framework. Water Resour. Res. 46, 1–19.
https://doi.org/10.1029/2010WR009227
Truex, M.J., 2018. Groundwater and Vadose Zone Remedy Management Strategies at
DOE Sites, in: SERDP ESTCP Symposium. Washington, D.C.
Truex, M.J., Johnson, T.C., Strickland, C.E., Peterson, J.E., Hubbard, S.S., 2013a.
Monitoring Vadose Zone Desiccation with Geophysical Methods. Vadose Zo. J.
12, 0. https://doi.org/10.2136/vzj2012.0147
Truex, M.J., Johnson, T.C., Strickland, C.E., Peterson, J.E., Hubbard, S.S., 2013b.
Monitoring Vadose Zone Desiccation with Geophysical Methods. Vadose Zo. J.
12. https://doi.org/10.2136/vzj2012.0147
Truex, M.J., Oostrom, M., Strickland, C.E., Chronister, G.B., Benecke, M.W., Johnson,
C.D., 2012. Field-Scale Assessment of Desiccation Implementation for Deep
Vadose Zone Contaminants. Vadose Zo. J. 11, 0.
Bibliography
293
https://doi.org/10.2136/vzj2011.0144
Tsai, F.T.-C., Elshall, A.S., 2013. Hierarchical Bayesian model averaging for
hydrostratigraphic modeling: Uncertainty segregation and comparative
evaluation. Water Resour. Res. 49, 5520–5536. https://doi.org/10.1002/wrcr.20428
Tsang, C.-F., Neretnieks, I., Tsang, Y., 2015. Hydrologic issues associated with nuclear
waste repositories. Water Resour. Res. 51, 1–50.
https://doi.org/10.1002/2014WR016259
Tso, C.-H.M., Kuras, O., Binley, A., 2019. On the Field Estimation of Moisture Content
Using Electrical Geophysics: The Impact of Petrophysical Model Uncertainty.
Water Resour. Res. 55, 7196–7211. https://doi.org/10.1029/2019WR024964
Tso, C.-H.M., Kuras, O., Wilkinson, P.B., Uhlemann, S., Chambers, J.E., Meldrum, P.I.,
Graham, J., Sherlock, E.F., Binley, A., 2017. Improved characterisation and
modelling of measurement errors in electrical resistivity tomography (ERT)
surveys. J. Appl. Geophys. 146, 103–119.
https://doi.org/10.1016/J.JAPPGEO.2017.09.009
Tso, C.-H.M., Zha, Y., Yeh, T.-C.J., Wen, J.-C., 2016. The relative importance of head,
flux, and prior information in hydraulic tomography analysis. Water Resour. Res.
52, 3–20. https://doi.org/10.1002/2015WR017191
Turkeltaub, T., Jia, X., Zhu, Y., Shao, M.-A., Binley, A., 2018. Recharge and Nitrate
Transport Through the Deep Vadose Zone of the Loess Plateau: A Regional-Scale
Model Investigation. Water Resour. Res. 54, 4332–4346.
https://doi.org/10.1029/2017WR022190
Uhlemann, S., Chambers, J., Wilkinson, P., Maurer, H., Merritt, A., Meldrum, P., Kuras,
O., Gunn, D., Smith, A., Dijkstra, T., 2017. Four-dimensional imaging of moisture
dynamics during landslide reactivation. J. Geophys. Res. Earth Surf. 122, 398–418.
https://doi.org/10.1002/2016JF003983
Uhlemann, S., Sorensen, J.P.R., House, A.R., Wilkinson, P.B., Roberts, C., Gooddy, D.C.,
2016. Integrated time-lapse geoelectrical imaging of wetland hydrological
processes. Water Resour. Res. 52, 1607–1625.
https://doi.org/10.1002/2015WR017932
Uhlemann, S., Wilkinson, P.B., Maurer, H., Wagner, F.M., Johnson, T.C., Chambers,
J.E., 2018. Optimized survey design for Electrical Resistivity Tomography:
combined optimization of measurement configuration and electrode placement.
Geophys. J. Int.
Varouchakis, E.A., Palogos, I., Karatzas, G.P., 2016. Application of Bayesian and cost
benefit risk analysis in water resources management 534, 390–396.
Vauhkonen, M., Karjalainen, P.A., Kaipio, J.P., 1998. A Kalman filter approach to track
fast impedance changes in electrical impedance tomography. IEEE Trans Biomed
Eng 45, 486–493. https://doi.org/10.1109/10.664204
Bibliography
294
Vereecken, H., Binley, A., Cassiani, G., Revil, A., Titov, K. (Eds.), 2006. Applied
Hydrogeophysics, NATO Science Series. Springer Netherlands, Dordrecht.
https://doi.org/10.1007/978-1-4020-4912-5
von Hebel, C., Rudolph, S., Mester, A., Huisman, J.A., Kumbhar, P., Vereecken, H., van
der Kruk, J., 2014. Three-dimensional imaging of subsurface structural patterns
using quantitative large-scale multiconfiguration electromagnetic induction data.
Water Resour. Res. 50, 2732–2748. https://doi.org/10.1002/2013WR014864
Vrugt, J.A., Robinson, B.A., 2007. Improved evolutionary optimization from
genetically adaptive multimethod search. Proc. Natl. Acad. Sci. 104, 708–711.
https://doi.org/10.1073/pnas.0610471104
Vrugt, J.A., Robinson, B.A., Hyman, J.M., 2009. Self-Adaptive Multimethod Search for
Global Optimization in Real-Parameter Spaces. IEEE Trans. Evol. Comput. 13,
243–259. https://doi.org/10.1109/TEVC.2008.924428
Wagner, B.J., 1992. Simultaneous parameter estimation and contaminant source
characterization for coupled groundwater flow and contaminant transport
modelling. J. Hydrol. 135, 275–303. https://doi.org/10.1016/0022-1694(92)90092-A
Wagner, F.M., 2016. New developments in electrical resistivity imaging with
applications to geological CO2 storage. ETH-Zürich. https://doi.org/10.3929/ethz-
a-010636965
Wagner, F.M., Bergmann, P., Rücker, C., Wiese, B., Labitzke, T., Schmidt-Hattenberger,
C., Maurer, H., 2015a. Impact and mitigation of borehole related effects in
permanent crosshole resistivity imaging: An example from the Ketzin CO2
storage site. J. Appl. Geophys. 123, 102–111.
https://doi.org/10.1016/j.jappgeo.2015.10.005
Wagner, F.M., Günther, T., Schmidt-Hattenberger, C., Maurer, H., 2015b. Constructive
optimization of electrode locations for target-focused resistivity monitoring.
Geophysics 80, E29–E40. https://doi.org/10.1190/geo2014-0214.1
Wagner, F.M., Wiese, B.U., 2018. Fully coupled inversion on a multi-physical reservoir
model – Part II: The Ketzin CO2 storage reservoir. Int. J. Greenh. Gas Control 75,
273–281. https://doi.org/10.1016/J.IJGGC.2018.04.009
Wainwright, H.M., Chen, J., Sassen, D.S., Hubbard, S.S., 2014a. Bayesian hierarchical
approach and geophysical data sets for estimation of reactive facies over plume
scales. Water Resour. Res. 50, 4564–4584. https://doi.org/10.1002/2013WR013842
Wainwright, H.M., Finsterle, S., Jung, Y., Zhou, Q., Birkholzer, J.T., 2014b. Making
sense of global sensitivity analyses. Comput. Geosci. 65, 84–94.
https://doi.org/10.1016/j.cageo.2013.06.006
Wainwright, H.M., Flores Orozco, A., Bücker, M., Dafflon, B., Chen, J., Hubbard, S.S.,
Williams, K.H., 2016. Hierarchical Bayesian method for mapping biogeochemical
hot spots using induced polarization imaging. Water Resour. Res. 52, 533–551.
Bibliography
295
https://doi.org/10.1002/2015WR017763
Wallin, E.L., Johnson, T.C., Greenwood, W.J., Zachara, J.M., 2013. Imaging high stage
river-water intrusion into a contaminated aquifer along a major river corridor
using 2-D time-lapse surface electrical resistivity tomography. Water Resour. Res.
49, 1693–1708. https://doi.org/10.1002/wrcr.20119
Wang, J., Chen, L., Su, R., Zhao, X., 2018. The Beishan underground research laboratory
for geological disposal of high-level radioactive waste in China: Planning, site
selection, site characterization and in situ tests. J. Rock Mech. Geotech. Eng. 10,
411–435. https://doi.org/10.1016/j.jrmge.2018.03.002
Ward, W.O.C., 2018. Development of machine learning techniques for characterising
changes in time-lapse resistivity monitoring. University of Nottingham.
Ward, W.O.C., Wilkinson, P.B., Chambers, J.E., Nilsson, H., Kuras, O., Bai, L., 2016.
Tracking tracer motion in a 4-D electrical resistivity tomography experiment.
Water Resour. Res. 52, 4078–4094. https://doi.org/10.1002/2015WR017958
Ward, W.O.C., Wilkinson, P.B., Chambers, J.E., Oxby, L.S., Bai, L., 2014. Distribution-
based fuzzy clustering of electrical resistivity tomography images for interface
detection. Geophys. J. Int. 197, 310–321. https://doi.org/10.1093/gji/ggu006
Watson, D.B., Doll, W.E., Gamey, T.J., Sheehan, J.R., Jardine, P.M., 2005. Plume and
lithologic profiling with surface resistivity and seismic tomography. Ground
Water 43, 169–77. https://doi.org/10.1111/j.1745-6584.2005.0017.x
Webster, D.S., Proctor, J.F., Marine, I.W., 1970. Two -Well Tracer Test in Fractured
Crystalline Two-Well Tracer Test in Fractured Crystalline. Washington DC.
Wehrer, M., Slater, L.D., 2015. Characterization of water content dynamics and tracer
breakthrough by 3-D electrical resistivity tomography (ERT) under transient
unsaturated conditions. Water Resour. Res. 51, 97–124.
https://doi.org/10.1002/2014WR016131
Weller, A., Slater, L., Nordsiek, S., 2013. On the relationship between induced
polarization and surface conductivity: Implications for petrophysical
interpretation of electrical measurements. Geophysics 78, D315–D325.
https://doi.org/10.1190/geo2013-0076.1
West, B., Welch, K.E., Galecki, A.T., 2007. Linear Mixed Models: a practical guide using
statistical software, 2nd ed. Chapman and Hall/CRC.
Whalley, W.R., Binley, A., Watts, C.W., Shanahan, P., Dodd, I.C., Ober, E.S., Ashton,
R.W., Webster, C.P., White, R.P., Hawkesford, M.J., 2017. Methods to estimate
changes in soil water for phenotyping root activity in the field. Plant Soil 415, 407–
422. https://doi.org/10.1007/s11104-016-3161-1
Whiteley, J.S., Chambers, J.E., Uhlemann, S., Wilkinson, P.B., Kendall, J.M., 2019.
Geophysical Monitoring of Moisture‐Induced Landslides: A Review. Rev.
Geophys. 57, 106–145. https://doi.org/10.1029/2018RG000603
Bibliography
296
Wiese, B., Wagner, F., Norden, B., Maurer, H., Schmidt-Hattenberger, C., 2018. Fully
coupled inversion on a multi-physical reservoir model - Part I: Theory and
concept. Int. J. Greenh. Gas Control.
Wilkinson, P., Chambers, J., Uhlemann, S., Meldrum, P., Smith, A., Dixon, N., Loke,
M.H., 2016. Reconstruction of landslide movements by inversion of 4-D electrical
resistivity tomography monitoring data. Geophys. Res. Lett. 43, 1166–1174.
https://doi.org/10.1002/2015GL067494
Wilkinson, P.B., Chambers, J.E., Lelliott, M., Wealthall, G.P., Ogilvy, R.D., 2008.
Extreme sensitivity of crosshole electrical resistivity tomography measurements
to geometric errors. Geophys. J. Int. 173, 49–62. https://doi.org/10.1111/j.1365-
246X.2008.03725.x
Wilkinson, P.B., Chambers, J.E., Meldrum, P.I., Gunn, D.A., Ogilvy, R.D., Kuras, O.,
2010a. Predicting the movements of permanently installed electrodes on an active
landslide using time-lapse geoelectrical resistivity data only. Geophys. J. Int. 183,
543–556. https://doi.org/10.1111/j.1365-246X.2010.04760.x
Wilkinson, P.B., Loke, M.H., Meldrum, P.I., Chambers, J.E., Kuras, O., Gunn, D.A.,
Ogilvy, R.D., 2012. Practical aspects of applied optimized survey design for
electrical resistivity tomography. Geophys. J. Int. 189, 428–440.
https://doi.org/10.1111/j.1365-246X.2012.05372.x
Wilkinson, P.B., Meldrum, P.I., Chambers, J.E., Kuras, O., Ogilvy, R.D., 2006. Improved
strategies for the automatic selection of optimized sets of electrical resistivity
tomography measurement configurations. Geophys. J. Int. 167, 1119–1126.
https://doi.org/10.1111/j.1365-246X.2006.03196.x
Wilkinson, P.B., Meldrum, P.I., Kuras, O., Chambers, J.E., Holyoake, S.J., Ogilvy, R.D.,
2010b. High-resolution Electrical Resistivity Tomography monitoring of a tracer
test in a confined aquifer. J. Appl. Geophys. 70, 268–276.
https://doi.org/10.1016/j.jappgeo.2009.08.001
Wilkinson, P.B., Uhlemann, S., Meldrum, P.I., Chambers, J.E., Carrière, S., Oxby, L.S.,
Loke, M.H.H., 2015. Adaptive time-lapse optimized survey design for electrical
resistivity tomography monitoring. Geophys. J. Int. 203, 755–766.
https://doi.org/10.1093/gji/ggv329
Williams, B.A., Thompson, M.D., Miller, S.F., 2012a. Integrated surface geophysical
investigation results at Liquid Efflucent Retention Facility, 200 East Area,
Washington. Richland, Washington.
Williams, B.A., Thompson, M.D., Miller, S.F., 2012b. Interpretation and integration of
seismic data in the Gable Gap. Richland, Washington.
Williams, B.A., Thompson, M.D., Miller, S.F., 2012c. Land streamer and gimbaled
geophones phase II-200 Areas: high-resolution seismic reflection suvey at the
Hanford Site. Richland, Washington.
Bibliography
297
Williams, B.A., Thompson, M.D., Miller, S.F., 2012d. Seismic reflection investigation at
the Liquid Effluent Retention Facility, 200 East Area, Hanford Site Richland,
Washington. Richland, Washington.
Williams, K.H., Kemna, A., Wilkins, M.J., Druhan, J., Arntzen, E., N’Guessan, A.L.,
Long, P.E., Hubbard, S.S., Banfield, J.F., 2009. Geophysical Monitoring of Coupled
Microbial and Geochemical Processes During Stimulated Subsurface
Bioremediation. Environ. Sci. Technol. 43, 6717–6723.
https://doi.org/10.1021/es900855j
Williams, M.D., Rockhold, M.L., Thorne, P.D., Chen, Y., 2008. Three-dimensional
groundwater models of the 300 Area at the Hanford Site, Washington State.
Richland, Washington.
Winship, P., Binley, A., Gomez, D., 2006. Flow and transport in the unsaturated
Sherwood Sandstone: characterization using cross-borehole geophysical methods.
Geol. Soc. London, Spec. Publ. 263, 219–231.
https://doi.org/10.1144/GSL.SP.2006.263.01.12
Wu, W.M., Carley, J., Fienen, M., Mehlhorn, T., Lowe, K., Nyman, J., Luo, J., Gentile,
M.E., Rajan, R., Wagner, D., Hickey, R.F., Gu, B., Watson, D., Cirpka, O. a.,
Kitanidis, P.K., Jardine, P.M., Criddle, C.S., 2006. Pilot-scale in situ bioremediation
of uranium in a highly contaminated aquifer. 1. Conditioning of a treatment zone.
Environ. Sci. Technol. 40, 3978–3985. https://doi.org/10.1021/es051954y
Xu, T., Gómez-Hernández, J.J., 2018. Simultaneous identification of a contaminant
source and hydraulic conductivity via the restart normal-score ensemble Kalman
filter. Adv. Water Resour. 112, 106–123.
https://doi.org/10.1016/j.advwatres.2017.12.011
Xu, T., Gómez-Hernández, J.J., 2016. Joint identification of contaminant source location,
initial release time, and initial solute concentration in an aquifer via ensemble
Kalman filtering. Water Resour. Res. 52, 6587–6595.
https://doi.org/10.1002/2016WR019111
Xue, L., Zhang, D., 2014. A multimodel data assimilation framework via the ensemble
Kalman filter. Water Resour. Res. 50, 4197–4219.
https://doi.org/10.1002/2013WR014525
Yamakawa, Y., Kosugi, K., Katsura, S., Masaoka, N., Mizuyama, T., 2012. Spatial and
Temporal Monitoring of Water Content in Weathered Granitic Bedrock Using
Electrical Resistivity Imaging. Vadose Zo. J. 11, 0.
https://doi.org/10.2136/vzj2011.0029
Yang, X., Chen, X., Carrigan, C.R., Ramirez, A.L., 2014. Uncertainty quantification of
CO2 saturation estimated from electrical resistance tomography data at the
Cranfield site. Int. J. Greenh. Gas Control 27, 59–68.
https://doi.org/10.1016/j.ijggc.2014.05.006
Yang, X., Lassen, R.N., Jensen, K.H., Looms, M.C., 2015. Monitoring CO2 migration in
Bibliography
298
a shallow sand aquifer using 3D crosshole electrical resistivity tomography. Int. J.
Greenh. Gas Control 42, 534–544. https://doi.org/10.1016/J.IJGGC.2015.09.005
Yeh, T.-C.J., Liu, S., 2000. Hydraulic tomography : Development of a new aquifer test
method. Water Resour. Res. 36, 2095–2105.
Yeh, T.-C.J., Liu, S., Glass, R.J., Baker, K., Brainard, J.R., Alumbaugh, D., LaBrecque, D.,
2002. A geostatistically based inverse model for electrical resistivity surveys and
its applications to vadose zone hydrology. Water Resour. Res. 38, WR001204.
https://doi.org/10.1029/2001WR001204
Zahner, T., Lochbühler, T., Mariethoz, G., Linde, N., 2016. Image synthesis with graph
cuts: a fast model proposal mechanism in probabilistic inversion. Geophys. J. Int.
204, 1179–1190. https://doi.org/10.1093/gji/ggv517
Zeng, L., Shi, L., Zhang, D., Wu, L., 2012. A sparse grid based Bayesian method for
contaminant source identification. Adv. Water Resour. 37, 1–9.
https://doi.org/10.1016/J.ADVWATRES.2011.09.011
Zha, Y., Yeh, T.-C.J., Illman, W.A., Mok, C.M.W., Tso, C.-H.M., Carrera, B.A., Wang,
Y.-L., 2019. Exploitation of pump-and-treat remediation systems for
characterization of hydraulic heterogeneity. J. Hydrol. 573, 324–340.
https://doi.org/10.1016/j.jhydrol.2019.03.089
Zhang, Y.-K., Neuman, S.P., 1990. A quasi-linear theory of non-Fickian and Fickian
subsurface dispersion: 2. Application to anisotropic media and the Borden site.
Water Resour. Res. 26, 903–913. https://doi.org/10.1029/WR026i005p00903
Zhang, Y., Liu, N., Oliver, D.S., 2010. Ensemble filter methods with perturbed
observations applied to nonlinear problems. Comput. Geosci. 14, 249–261.
https://doi.org/10.1007/s10596-009-9149-7
Zhang, Y., Schaap, M.G., Guadagnini, A., Neuman, S.P., 2016. Inverse modeling of
unsaturated flow using clusters of soil texture and pedotransfer functions. Water
Resour. Res. 52, 7631–7644. https://doi.org/10.1002/2016WR019016
Zhao, Z., 2000. Application of geophysical exploration to site selection for waste
disposal of the Guangdong nuclear power plant. Geol. Prospect. (in Chinese) 36,
51–53.
Zheng, J., Zhang, Z., Dong, L., 2000. Environmental geophysics and its advance. Adv.
Earth Sci. (in Chinese) 15, 40–47.
Zheng, Q., Zhang, J., Xu, W., Wu, L., Zeng, L., 2018. Adaptive Multi‐Fidelity Data
Assimilation for Nonlinear Subsurface Flow Problems. Water Resour. Res.
2018WR023615. https://doi.org/10.1029/2018WR023615
Zhou, B., Dahlin, T., 2003. Properties and effects of measurement errors on 2D
resistivity imaging surveying. Near Surf. Geophys. 1, 105–117.
https://doi.org/10.3997/1873-0604.2003001
Bibliography
299
Zhou, H., Gómez-Hernández, J.J., Li, L., 2014. Inverse methods in hydrogeology:
Evolution and recent trends. Adv. Water Resour. 63, 22–37.
https://doi.org/10.1016/j.advwatres.2013.10.014
Zhou, J., Revil, A., Jardani, A., 2016. Stochastic structure-constrained image-guided
inversion of geophysical data. Geophysics 81, E89–E101.
https://doi.org/10.1190/geo2014-0569.1
Zhou, J., Revil, A., Karaoulis, M., Hale, D., Doetsch, J., Cuttler, S., 2014. Image-guided
inversion of electrical resistivity data. Geophys. J. Int. 197, 292–309.
https://doi.org/10.1093/gji/ggu001
Zhu, P., Shi, L., Zhu, Y., Zhang, Q., Huang, K., Williams, M., 2017. Data assimilation of
soil water flow via ensemble Kalman filter: Infusing soil moisture data at different
scales. J. Hydrol. 555, 912–925. https://doi.org/10.1016/j.jhydrol.2017.10.078
Zovi, F., Camporese, M., Hendricks Franssen, H.J., Huisman, J.A., Salandin, P., 2017.
Identification of high-permeability subsurface structures with multiple point
geostatistics and normal score ensemble Kalman filter. J. Hydrol. 548, 208–224.
https://doi.org/10.1016/j.jhydrol.2017.02.056
Appendix 1: Instructions on using PFLOTRAN-E4D
300
Appendix 1: Instructions on using PFLOTRAN-E4D
A significant portion of this thesis uses the software PFLOTRAN-E4D (Johnson et al.,
2017) for hydrogeophysical simulations. The software had been part of the PFLOTRAN
releases but this feature is no longer supported by the end of 2018. The section
documents an alternative procedure to run hydrogeophysical simulation by calling
PFLOTRAN and E4D sequentially.
The approach documented in Johnson et al. (2017) runs PFLOTRAN and E4D in
parallel by assigning a number of nodes to run the former and others to run the latter.
When optimized, this can be advantageous in terms of run time, especially for very
long hydrological simulations. An alternative and perhaps more straightforward and
flexible approach is to run PFLOTRAN and then run E4D (each in parallel using all
available resources). We can extract outputs from the PFLOTRAN output file (in HDF5
format) using a script, call the FORTRAN mapping subroutines in Johnson et al. (2017),
and write to E4D input files. The scripts are available from the thesis author.
Steps (on a Linux-type machine):
1. Install PFLOTRAN and E4D (if not already)
2. Install f2py so that FORTRAN codes can be importable to python (if not
already). After installation, load python in your terminal and type:
f2py -c -m test_interp2 test_interp2.f90
f2py -c -m mapit mapit.f90
Make sure the resultant executables are in the directory to run PFLOTRAN-E4D.
3. Now you can call two of the python scripts in the folder:
build_pf_to_e4d_mat
rix.py
It builds the interpolation matrix. You only need to
call it once for each set of PFLOTRAN and E4D
meshes.
map_pf_to_e4d.py It maps PFLOTRAN outputs to E4D mesh and
perform petrophysical transform (You can modify it
Appendix 1: Instructions on using PFLOTRAN-E4D
301
to specify whatever model you like). It needs to be
called whenever E4D needs to be run. Make sure
you review the script and understand the mapping
it uses.
You will also need to change these files, which are read by the scripts.
sim_times.txt
(or equivalent)
Specifies the PFLOTRAN output times to which E4D is
called (note: they must be specified in the PFLTORAN
input file), the background conductivity file, and the
petrophysical transform file.
background.si
g
Specifies electrical conductivity for each cell in the E4D
mesh. This value will be used if interpolation from
PFLOTRAN is not available.
wax_smit.sig Specifies petrophysical parameters (I.e. Waxman-Smit here)
for each cell in the E4D mesh
All files required
to run E4D mode 2
(forward model)
An example bash script for running PFLOTRAN-E4D this way
(hydrogeophysics.sh):
#!/bin/bash -l ## Do not comment out this line. Must be
included for bash scripts
module load python
## run PFLOTRAN, for example:
$ mpirun -n 4 $PFLOTRAN_DIR/src/pflotran/pflotran -pflotranin
myinputfile.in
Appendix 1: Instructions on using PFLOTRAN-E4D
302
## build interpolation matrix
python build_pf_to_e4d_matrix.py
## map PF output to sigma files and run E4D for each specified
times.
python map_pf_to_e4d.py
Notes on the PFLOTRAN (.in) input file in order to run PFLOTRAN-E4D
1. The output time which E4D will call must be specified as PFLOTRAN output
times
2. “Tracer” must be specified as a primary species
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method
303
Appendix 2: A guide to performing global sensitivity analysis
using the Morris (1991) method
The following is a exrept of an interactive Jupyter Notebook with python codes
included in the ResIPy distribution (link).
Global sensitivity analysis is a Monte Carlo based method to rank the importance of
parameters in a given modelling problem. As opposed to local senstivity analysis, it
does not require the construction of the Jacobian, making it a flexible tool to evaluate
complex problems.
Global sensitivty analysis is available in mainly uncertainty quantificaiton packages,
as well as some flow and transport programs (e.g. iTOUGH2). GSA is also very
popular in catchment modelling and civil engineering/risk analysis problems.
Some GSA work in hydrogeophysics (mainly by Berkeley Lab):
coupled hydrological-thermal-geophysical inversion (Tran et al 2017)
making sense of global senstivity analysis (Wainwright et al 2014)
Sensitivity analysis of environmental models (Pianosi et al 2014)
hydrogeology of a nuclear site in the Paris Basin (Deman et al 2016)
Global Sensitivity and Data-Worth Analyses in iTOUGH2 User's Guide
(Wainwright et al 2016)
In this tutorial, we will see how to link the ResIPy API and SALib for senstivity
analysis. Two key elements of SA are (i) forward modelling (Monte Carlo runs) and (ii)
specifying the parameter ranges. This notebook will showcase of the use of the Method
of Morris, which is known for its relatively small computational cost. This tutorial is
modified from the one posted on https://github.com/SALib/SATut to demonstrate its
coupling with ResIPy
Morris sensitivity method
The Morris one-at-a-time (OAT) method (Morris, 1991) can be considered as an
extension of the local sensitivity method. Each parameter range is scaled to the unit
interval [0, 1] and partitioned into (p−1)(p−1) equally-sized intervals. The reference
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method
304
value of each parameter is selected randomly from the set 0,1/(p−1),2/(p−1),…,1− Δ.
The fixed increment Δ=p/2(p−1) is added to each parameter in random order to
compute the elementary effect (EE) of 𝑥𝑖
𝐸𝐸𝑖 =1
𝜏𝑦
𝑓(𝑥1 ∗,… , 𝑥𝑖 ∗ +𝛥,… , 𝑥𝑘 ∗) − 𝑓(𝑥1 ∗, … , 𝑥𝑘 ∗)
𝛥
where 𝑥𝑖 ∗ is the randomly selected parameter set, and τy is the output-scaling factor.
To compute EEi for k parameters, we need (𝑘 + 1) simulations (called one “path”) in
the same way as that of the local sensitivity method. By having multiple paths, we have
an ensemble of EEs for each parameter. The total number of simulations is 𝑟(𝑘 + 1),
where r is the number of paths.
We compute three statistics: the mean 𝐸𝐸, standard deviation (STD) of 𝐸𝐸, and mean
of absolute 𝐸𝐸.
mean EE (μ) represents the average effect of each parameter over the parameter space,
the mean EE can be regarded as a global sensitivity measure.
mean |EE| (μ∗) is used to identify the non-influential factors,
STD of EE (σ) is used to identify nonlinear and/or interaction effects. (The standard
error of mean (SEM) of EE, defined as 𝑆𝐸𝑀 = 𝑆𝑇𝐷/𝑟0.5 , is used to calculate the
confidence interval of mean EE (Morris, 1991))
Importing libraries
%matplotlib inline
import warnings
warnings.filterwarnings('ignore')
import os
import sys
sys.path.append((os.path.relpath('../src'))) # add here the relative
path of the API folder
import numpy as np # numpy for electrode generation
import pandas as pd
from IPython.utils import io # suppress R2 outputs during MC runs
from resipy.R2 import R2
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method
305
API path = C:\Users\mtso\Downloads\pyr2-master\src\resipy
ResIPy version = 1.1.6
The SALib package
SALib is a free open-source Python library
If you use Python, you can install it by running the command
pip install SALib
Documentation is available online and you can also view the code on Github.
The library includes:
Sobol Sensitivity Analysis (Sobol 2001, Saltelli 2002, Saltelli et al. 2010)
Method of Morris, including groups and optimal trajectories (Morris
1991, Campolongo et al. 2007)
Fourier Amplitude Sensitivity Test (FAST) (Cukier et al. 1973, Saltelli et al. 1999)
Delta Moment-Independent Measure (Borgonovo 2007, Plischke et al. 2013)
Derivative-based Global Sensitivity Measure (DGSM) (Sobol and Kucherenko
2009)
Fractional Factorial Sensitivity Analysis (Saltelli et al. 2008)
SALib Tutorial
# import the packages
from SALib.sample import morris as ms
from SALib.analyze import morris as ma
from SALib.plotting import morris as mp
Create ERT forward problem with ResIPy
In the code below, created a R2 forward problem to be analyzed
k = R2()
elec = np.zeros((24,3))
elec[:,0] = np.arange(0, 24*0.5, 0.5) # with 0.5 m spacing and 24 ele
ctrodes
k.setElec(elec)
#print(k.elec)
# defining electrode array
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method
306
x = np.zeros((24, 3))
x[:,0] = np.arange(0, 24*0.5, 0.5)
k.setElec(elec)
# creating mesh
k.createMesh(res0=20)
# add region
k.addRegion(np.array([[2,-0.3],[2,-2],[3,-2],[3,-0.3],[2,-0.3]]), 50)
k.addRegion(np.array([[5,-2],[5,-3.5],[8,-3.5],[8,-2],[5,-2]]), 500)
# define sequence
k.createSequence([('dpdp1', 1, 10)])
# forward modelling
k.forward(noise=0.025)
# read results
fwd_dir = os.path.relpath('../src/resipy/invdir/fwd')
obs_data = np.loadtxt(os.path.join(fwd_dir, 'R2_forward.dat'),skiprow
s =1)
obs_data = obs_data[:,6]
# plot
k.showMesh()
Working directory is: C:\Users\mtso\Downloads\pyr2-master\src\resipy\
invdir
clearing the dirname
computed DOI : -7.67
Using a quadrilateral mesh.
quad
written mesh.dat file to
C:\Users\mtso\Downloads\pyr2-master\src\resipy\invdir\mesh.dat
Writing .in file...
done
Writing protocol.dat ...
done
Running forward model
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method
307
>> R 2 R e s i s t i v i t y I n v e r s i o n v3.3 <<
>> D a t e : 21 - 08 - 2019
>> My beautiful survey
>> F o r w a r d S o l u t i o n S e l e c t e d <<
>> Determining storage needed for finite element conductance matrix
>> Generating index array for finite element conductance matrix
>> Reading start resistivity from resistivity.dat
Measurements read: 165 Measurements rejected: 0
>> Total Memory required is: 0.395 Gb
Inf or NaN: filterData: 0 / 165 quadrupoles removed.
strange quadrupoles: filterData: 0 / 165 quadrupoles removed.
165/165 reciprocal measurements NOT found.
0 measurements error > 20 %
computed DOI : -3.67
Forward modelling done.
Mesh plotted in 0.08826 seconds
Define a problem file
In the code below, a problem file is used to define the parameters and their ranges we
wish to explore, which corresponds to the following table:
Parameter Range Description
rho0 [ohm m] 10^[0.5,3.5] background
rho1 [ohm m] 10^[0.5,3.5] inclusion A
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method
308
Parameter Range Description
rho2 [ohm m] 10^[0.5,3.5] inclusion B
morris_problem =
'num_vars': 3,
# These are their names
'names': ['rho1', 'rho2', 'rho3'], # can add z1 z2 etc.
# Plausible ranges over which we'll move the variables
'bounds': [[0.5,3.5], # log10 of rho (ohm m)
[0.5,3.5],
[0.5,3.5]#,
# [-3,-1],
# [-7,-4],
],
# I don't want to group any of these variables together
'groups': None
Generate a Sample
We then generate a sample using the morris.sample() procedure from the SALib
package.
number_of_trajectories = 20
sample = ms.sample(morris_problem, number_of_trajectories, num_levels
=10)
len(sample)
print(sample[79,:])
[1.83333333 1.83333333 0.5 ]
Run the sample through the monte carlo procedure in R2
Great! You have defined your problem and have created a series of input files for forward
runs. Now you need to run R2 for each of them to obtain their ERT responses.
For this example, each sample takes a few seconds to run on a PC.
#%%capture
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method
309
simu_ensemble = np.zeros((len(obs_data),len(sample)))
for ii in range(0, len(sample)):
with io.capture_output() as captured: # suppress inline
output from ResIPy
# creating mesh
k.createMesh(res0=10**sample[ii,0]) # need to use more effe
ctive method, no need to create mesh every time
# add region
k.addRegion(np.array([[2,-0.3],[2,-2],[3,-2],[3,-0.3],[2,-0.3
]]), 10**sample[ii,1])
k.addRegion(np.array([[5,-2],[5,-3.5],[8,-3.5],[8,-2],[5,-2]]
), 10**sample[ii,2])
# forward modelling
k.forward(noise=0.025, iplot = False)
out_data = np.loadtxt(os.path.join(fwd_dir, 'R2_forward.dat')
,skiprows =1)
simu_ensemble[:,ii] = out_data[:,6]
print("Running sample",ii+1)
Running sample 1
Running sample 2
Running sample 3
…
Running sample 77
Running sample 78
Running sample 79
Running sample 80
Factor Prioritisation
We'll run a sensitivity analysis of the power module to see which is the most influential
parameter.
The results parameters are called mu, sigma and mu_star.
Mu is the mean effect caused by the input parameter being moved over its range.
Sigma is the standard deviation of the mean effect.
Mu_star is the mean absolute effect.
The higher the mean absolute effect for a parameter, the more sensitive/important it is*
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method
310
# Define an objective function: here I use the error weighted rmse
def obj_fun(sim,obs,noise):
y = np.divide(sim-obs,noise) # weighted data misfit
y = np.sqrt(np.inner(y,y))
return y
output = np.zeros((1,len(sample)))
for ii in range(0, len(sample)):
output[0,ii] = obj_fun(simu_ensemble[:,ii],obs_data,0.025*obs_dat
a) # assume 2.5% noise in the data
# Store the results for plotting of the analysis
Si = ma.analyze(morris_problem, sample, output, print_to_console=Fals
e)
print(":20s :>7s :>7s :>7s".format("Name", "mean(EE)", "mean(
|EE|)", "std(EE)"))
for name, s1, st, mean in zip(morris_problem['names'],
Si['mu'],
Si['mu_star'],
Si['sigma']):
print(":20s :=7.3f :=7.3f :=7.3f".format(name, s1, st, me
an))
Name mean(EE) mean(|EE|) std(EE)
rho1 78050.042 78050.042 38362.374
rho2 2595.171 2738.277 4484.274
rho3 1594.958 1595.198 2568.473
# make a plot
import matplotlib.pyplot as plt
import numpy as np
fig, ax = plt.subplots()
ax.scatter(Si['mu_star'],Si['sigma'])
ax.plot(Si['mu_star'],2*Si['sigma']/np.sqrt(number_of_trajectories),'
--',alpha=0.5)
ax.plot(np.array([0,Si['mu_star'][0]]),2*np.array([0,Si['sigma'][0]/n
p.sqrt(number_of_trajectories)]),'--',alpha=0.5)
plt.title('Distribution of Elementary effects')
plt.xlabel('mean(|EE|)')
plt.ylabel('std($EE$)')
Appendix 2: A guide to performing global sensitivity analysis using the Morris (1991)
method
311
for i, txt in enumerate(Si['names']):
ax.annotate(txt, (Si['mu_star'][i], Si['sigma'][i]))
# higher mean |EE|, more important factor
# line within the dashed envelope means nonlinear or interaction effe
cts dominant
Appendix 3: Annotated bibliography for related textbooks
312
Appendix 3: Annotated bibliography for related textbooks
1. Scheidt, C., Li, L., Caers, J., 2018. Quantifying Uncertainty in Subsurface
Systems, Geophysical Monograph Series. John Wiley & Sons, Inc., Hoboken, NJ,
USA.
This book (Scheidt et al., 2018) begins by stating its motivation—the earth
resources challenge and the challenge to make decision under uncertainty. It
then gives an excellent review of available data science tools that are relevant
for UQ and introduces sensitivity analysis methods and Bayesianism. It is
followed by a detailed review on geological priors and inversion, which is
rarely found in other text. Subsequently, it introduces the concept of Bayesian
evidential learning and provides several application example of UQ. Finally,
this book provides an overview of available computer codes and an outlook for
UQ in subsurface systems.
2. Sun, N.-Z., Sun, A., 2015. Model Calibration and Parameter Estimation.
Springer New York, New York, NY.
This book (Sun and Sun, 2015) begins with a review both classical multi-
objective and statistical parameter estimation methods. It is followed by a
review for model differentiation, model dimension reduction, and model
structure identification methods. Its final chapters reviewed goal-oriented
modelling, uncertainty quantification and optimal experimental design
methods for environmental inverse problems. The major strength of this books
is its completeness and it links classical and statistical inverse problem
formulation.
3. Eidsvik, J., Mukerji, T., and , Bhattacharjya, D. ,2015. Value of Information in
the Earth Sciences. Cambridge University Press, Cambridge.
This book (Eidsvik et al., 2015) reviews the concept of value of information (VoI)
analysis in subsurface characterisation, which is popular in petroleum and
mining industry but also sees growing applications in groundwater protection
and environmental conservation. It presents a unified framework for assessing
the value of potential data gathering schemes by integrating spatial modelling
and decision analysis, which is useful for site characterisation as it can be
Appendix 3: Annotated bibliography for related textbooks
313
applied to determine whether a proposed survey can provide sufficicient
information that justifies its cost. This book also describes relevant quantitative
tools such as decision trees and influence diagrams, as well as models for
continuous and discrete dependent spatial variables, including Bayesian
networks, Markov random fields, Gaussian processes, and multiple-point
geostatistics.
Vita
314
Vita
2007 Hong Kong Certicate of Education Examination: Queen’s College,
Hong Kong
2012 BS in geosystems engineering and hydrogeology, The University
of Texas at Austin, Austin (TX), USA
2012 – 2015 Graduate research assistant, University of Arizona, Tucson (AZ),
USA
2015 MS in hydrology, University of Arizona, Tucson (AZ), USA
2015 – 2019 PhD student, Lancaster Environment Centre, Lancaster
University, Lancaster, UK
2019 – Research associate in environmental data science, Centre for
Ecology and Hydrology, Lancaster, UK